Optimization method of TS model for gas turbine based on multi-model predictive control algorithm

Abstract

The present invention discloses a gas turbine T-S model optimization method based on a multi-model predictive control algorithm. The method comprises data preprocessing, using five adjacent sampling points to perform cubic polynomial fitting to preprocess the data and improve the data quality; establishing a characteristic vector, that is, constructing a characteristic vector combination of input and output variables; performing antecedent structure identification on the characteristic vector; performing consequent parameter identification on the data obtained in step S3; obtaining a T-S fuzzy model; obtaining a step response vector model of a corresponding working condition in the T-S fuzzy model obtained in step S5, and using a DMC algorithm to obtain an optimal input quantity to optimize the gas turbine output.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic operating conditions of gas turbines, and in particular to a gas turbine TS model optimization method based on a multi-model predictive control algorithm. Background Art

[0002] As a major energy country, my country is actively participating in addressing global climate change. The energy and power industry will be reshaped. Coal-fired power, which has long been the main energy source in my country, is transforming into a regulatory energy source. It has become an inevitable trend to drive clean energy production with renewable energy [1]. Wind and solar power generation is limited by many factors, and the uncertainty of output will have a serious impact on the stability regulation and flexibility scheduling of the power system. As a mainstream power generation energy source, new energy power generation cannot be completed overnight and still requires reasonable planning and time transition. During this period, gas-steam combined cycle units can both support system transformation and cooperate with the development of new energy, and are the best option under existing conditions.

[0003] Gas-steam combined cycle units belong to thermal power generation. Unlike traditional pulverized coal-fired units, the combustion system of gas-steam combined cycle units is completed by gas turbines, and the fuel selected is natural gas (methane CH4), which obtains a higher calorific value and zero pollution emissions after full combustion. The gas-steam combined cycle unit has a significant emission reduction effect and can provide a high proportion of effective capacity. In recent years, power outages have occurred in many parts of my country, reflecting the supply imbalance between the power generation side and the power grid side of the power system, and the mismatch between power supply capacity and maximum load. This is due to the growth of residents' consumption and changes in the social electricity consumption structure, especially in winter and summer, when the electricity load increases, the peak-to-valley difference of the power grid continues to expand, and the frequency and voltage regulation capacity is insufficient, highlighting the severity of the problem of power supply security. Therefore, the capacity of the newly added gas-steam combined cycle units can effectively respond to and alleviate the problem of power grid peak regulation.

[0004] However, the special working nature of gas turbines also brings difficulties in understanding their dynamic characteristics. In addition, due to the nonlinearity and coupling of gas turbines themselves, it is not a good choice to directly use mechanism methods to study gas turbines. In recent years, the trend and hot spot of studying nonlinear systems is to decompose nonlinear systems into multiple linear system models, and then fit each segment of linear system to the global nonlinear system by fusion. The most typical of this multi-model strategy is the TS clustering fuzzy model. There is no need to give a model of the nonlinear system. It is only necessary to study the relationship between the input and output data of the gas turbine, use fuzzy clustering and identification methods, divide the data space into multiple subspaces, and express each group of simple linear models in segments. Establishing an accurate and universal gas turbine model is the basis for optimization and control, which is of great practical significance. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention aims to provide a gas turbine TS model optimization method based on a multi-model predictive control algorithm.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A gas turbine TS model optimization method based on a multi-model predictive control algorithm, the method comprising:

[0008] S1 data preprocessing, using five adjacent sampling points to perform cubic polynomial fitting to preprocess the data and improve data quality;

[0009] S2 establishes the feature vector, that is, constructs the feature vector combination of input and output variables;

[0010] S3 identifies the antecedent structure of the feature vector;

[0011] S4 performs subsequent parameter identification on the data obtained in step S3;

[0012] S5 obtains the TS fuzzy model;

[0013] S6 obtains a step response vector model of the corresponding working condition in the TS fuzzy model obtained in step S5;

[0014] S7 uses the DMC algorithm to find the optimal input to optimize the gas turbine output.

[0015] It should be noted that in step S1, a five-point cubic smoothing formula is used for data preprocessing, where:

[0016] Let the original data be x, and the filtered data be y. The specific filtering algorithm rules are expressed as:

[0017]

[0018] It should be noted that the feature vector combination in step S2 is:

[0019]

[0020] Among them, X1 represents the characteristic vector combination of power, k represents the kth sampling time, which can also be understood as the order in the discrete system, X2 represents the characteristic vector combination of exhaust temperature, u1 represents the input variable fuel amount, u2 represents the input variable inlet guide vane opening, y1 represents the output variable engine power, and y2 represents the output variable exhaust temperature.

[0021] It should be noted that the step S3 adopts an improved FCM clustering algorithm, specifically:

[0022]

[0023] in, γ is the regularization factor.

[0024] It should be noted that it also includes an improved particle swarm algorithm, specifically:

[0025] The first step is to define the variables to be identified, set the optimization objective function and stop condition, randomly generate a population of several particles, and initialize the position and velocity of the particles.

[0026] In the second step, the numerical value initially identified by Matlab is used as a reference to constrain the interval of particle search for optimal solution and speed up the search for the optimal solution;

[0027] The third step is to calculate the fitness value of each particle and record the current historical optimal solution and global optimal solution of each particle;

[0028] The fourth step is to determine the distance between each particle and the global optimal particle according to the formula, and calculate the inertia weight and learning factor of the particle respectively;

[0029] Step 5: Update the particle's velocity and position and record them, overwrite the original historical optimal solution and the global optimal solution, bring them into the objective function to solve, and determine whether the stop condition is met. If so, proceed to the next step, otherwise execute the third step;

[0030] Step 6: Output the globally optimal particle individual, i.e. the solution of the identified parameters;

[0031] The accurate transfer function of the gas turbine site is obtained by improving the particle swarm algorithm. The improved transfer formula is:

[0032]

[0033] The beneficial effects of the present invention are:

[0034] The present invention is based on the "decomposition-fusion" strategy and data-driven TS fuzzy identification modeling method, which simplifies the gas turbine into a two-input and two-output model. Furthermore, the model structure is identified by using an improved fuzzy C-means clustering (FCM) algorithm, and the model parameters are identified by using the least squares method with both result correction and recursive methods. A mathematical model with linear rule expression of fuzzy statements is obtained, and the accuracy and versatility of the model are verified by regression model indicators.

[0035] Secondly, the present invention combines the established gas turbine model with dynamic matrix predictive control and applies it to the output predictive control strategy. According to the results, the TS fuzzy model established by the present invention combined with the hybrid algorithm of predictive control has better control effect than PID control. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is the result of five-point cubic method smoothing of the input data in the present invention;

[0037] Figure 2 The five-point cubic method smoothing result of the output data of the present invention;

[0038] Figure 3 It is a graph showing the error function improved by different γ factors in the present invention;

[0039] Figure 4 This is a diagram of the identification result of the gas turbine TS model in the present invention;

[0040] Figure 5 It is a schematic diagram comparing the excitation and actual response under the transfer function of the gas turbine in the present invention;

[0041] Figure 6 It is a schematic diagram of the structure of the gas turbine output prediction control in the present invention;

[0042] Figure 7 It is a power setpoint disturbance control strategy diagram of the present invention;

[0043] Figure 8 It is a schematic diagram of the dynamic response of the power target value tracking experiment in the present invention;

[0044] Fig. 9 It is a control strategy structure diagram under external disturbance in the present invention;

[0045] Fig.10 It is a schematic diagram of the step response of the exhaust temperature disturbance signal in the simulation experiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described below in conjunction with the accompanying drawings. It should be noted that this embodiment is based on the technical solution and provides a detailed implementation method and a specific operation process, but the protection scope of the present invention is not limited to this embodiment.

[0047] The present invention is a gas turbine TS model optimization method based on a multi-model predictive control algorithm, the method comprising:

[0048] S1 data preprocessing, using five adjacent sampling points to perform cubic polynomial fitting to preprocess the data and improve data quality;

[0049] S2 establishes the feature vector, that is, constructs the feature vector combination of input and output variables;

[0050] S3 identifies the antecedent structure of the feature vector;

[0051] S4 performs subsequent parameter identification on the data obtained in step S3;

[0052] S5 obtains the TS fuzzy model;

[0053] S6 obtains a step response vector model of the corresponding working condition in the TS fuzzy model obtained in step S5;

[0054] S7 uses the DMC algorithm to find the optimal input to optimize the gas turbine output.

[0055] Furthermore, in step S1 of the present invention, a five-point cubic smoothing formula is used for data preprocessing, wherein:

[0056] Let the original data be x, and the filtered data be y. The specific filtering algorithm rules are expressed as:

[0057]

[0058] Furthermore, the feature vector combination in step S2 of the present invention is:

[0059]

[0060] Among them, X1 represents the characteristic vector combination of power, k represents the kth sampling time, which can also be understood as the order in the discrete system, X2 represents the characteristic vector combination of exhaust temperature, u1 represents the input variable fuel amount, u2 represents the input variable inlet guide vane opening, y1 represents the output variable engine power, and y2 represents the output variable exhaust temperature.

[0061] Furthermore, the step S3 of the present invention adopts an improved FCM clustering algorithm, specifically:

[0062]

[0063] in, γ is the regularization factor.

[0064] Furthermore, the present invention also includes an improved particle swarm algorithm, specifically:

[0065] The first step is to define the variables to be identified, set the optimization objective function and stop condition, randomly generate a population of several particles, and initialize the position and velocity of the particles.

[0066] In the second step, the numerical value initially identified by Matlab is used as a reference to constrain the interval of particle search for optimal solution and speed up the search for the optimal solution;

[0067] The third step is to calculate the fitness value of each particle and record the current historical optimal solution and global optimal solution of each particle;

[0068] The fourth step is to determine the distance between each particle and the global optimal particle according to the formula, and calculate the inertia weight and learning factor of the particle respectively;

[0069] Step 5: Update the particle's velocity and position and record them, overwrite the original historical optimal solution and the global optimal solution, bring them into the objective function to solve, and determine whether the stop condition is met. If so, proceed to the next step, otherwise execute the third step;

[0070] Step 6: Output the globally optimal particle individual, i.e. the solution of the identified parameters;

[0071] The accurate transfer function of the gas turbine site is obtained by improving the particle swarm algorithm. The improved transfer formula is:

[0072]

[0073] Example

[0074] It should be pointed out that, in view of the nonlinear characteristics of gas turbines, such as multiple input and output variables and strong coupling between variables, the use of data-driven system identification methods to establish a mathematical model that can accurately express the gas turbine when only input and output data are available is one of the current methods to solve the problem of nonlinear system modeling. This paper uses the TS fuzzy model identification algorithm to model the gas turbine, uses a linear fuzzy model to describe the dynamic characteristics of the nonlinear system, and then applies the optimization theory to the fuzzy model identification, establishes a gas turbine identification strategy process, and then obtains the TS fuzzy model of the gas turbine.

[0075] 1. Data Preprocessing

[0076] This paper uses the five-point cubic smoothing formula, that is, using five adjacent sampling points to perform cubic polynomial fitting to preprocess the data. The algorithm is simple and the processing effect is good. Let the original data be recorded as x, and the filtered data be recorded as The specific filtering algorithm rules are expressed as follows:

[0077]

[0078] The data in this article are taken from the DCS simulation machine. The gas-steam combined cycle unit was subjected to a load increase and decrease experiment. The experimental data of the unit was collected from the full load of 800MW to 700MW and then to 600MW. After the unit stabilized for a period of time, the unit was increased from 600MW to the full load of 800MW, and it was still in the 700MW stage in the middle. The rate of change of the unit load increase and decrease was set to 1, and the data sampling period was 1 second. The processing formula for the last two data points is similar to that for the first two points. Specifically, the processing of the first two sample points in practice is replaced by the third point. For the processing of the last two sample points, the five-point cubic smoothing formula is used to preprocess the two input variables and two output variables of the gas turbine. The results after processing are as follows: Figure 1 and Figure 2 shown.

[0079] 2. Establishing feature vectors

[0080] For systems with multiple input and output variables, there is coupling and nonlinearity between the data. In order to understand the impact of variables on the system at different times, it is necessary to analyze the input and output variables with the help of cross-correlation functions. The analysis index is to determine the correlation coefficient of the input variable to the system output at which time. The closer the correlation coefficient value is to 1, the greater the impact on the output. Analyzing the correlation coefficient can also avoid overfitting and improve the modeling accuracy of the model, so it is necessary to construct a feature vector combination of input and output variables.

[0081] After repeated experiments, the following formula was determined as the characteristic vector combination:

[0082]

[0083] Among them, X1 represents the characteristic vector combination of power, k represents the kth sampling time, which can also be understood as the order in the discrete system, X2 represents the characteristic vector combination of exhaust temperature, u1 represents the input variable fuel amount, u2 represents the input variable inlet guide vane opening, y1 represents the output variable engine power, and y2 represents the output variable exhaust temperature.

[0084] 3. Antecedent Identification—Improved FCM Clustering Algorithm

[0085] Fuzzy C-means clustering algorithm (FCM) is a flexible clustering algorithm. The core idea is to use data membership as the standard to determine the degree to which each data point belongs to a certain cluster and divide the data sample points into the interval [0,1]. First, multiple cluster centers are randomly selected, and the optimal target weighted point is searched according to the objective function. The fuzzy membership matrix and cluster center matrix are repeatedly adjusted to obtain the minimum number of iterations, and the optimal cluster center is selected. The clustering result is obtained through the fuzzy membership matrix U = [μ ij]∈R represents, U is the sample set, and finally the linear segments are connected by the membership function. The optimal solution of the objective function is solved by the Euclidean distance from the sample point to the cluster center. The objective function of the FCM algorithm can be expressed as follows:

[0086]

[0087] Since the FCM algorithm is highly dependent on the distance of sample points, outliers and noise points in the sample cluster will cause disturbances to the objective function, and it is easy to fall into local minima during the iteration process, affecting the final model accuracy. Therefore, a regularization factor γ is introduced to control the improved error function to reduce the impact of outliers and noise points and improve the role of effective sample points.

[0088] The original FCM objective function (D ij ) 2 Part of it is modified into the formula:

[0089]

[0090] like Figure 3 As shown, observe the left and right areas of the contour lines. When , the larger the value of γ is, the more obvious the suppression effect on noise points and outliers is; when When , the role of valid data points in the sample set can also weaken the impact of the error function.

[0091] 4. Post-item identification - recursive least squares parameter identification

[0092] The input variable matrix X obtained from the actual system is [x kj ], (k = 1, 2, ..., nj = 1, 2, ..., d) and the output vector y = [y1, y2, ..., y n ] T , where n is the number of samples and d is the dimension. Thus, the subsequent parameters of the model can be obtained from the following matrix equation:

[0093]

[0094] in is the error vector, Φ=[φ1,φ2,...,φ r ], (r = c × (d + 1)) is the coefficient matrix. From the previous description, we can see the formula:

[0095] φ i =[φ i (x1),φ i (x2), ..., φ i (x n )], (i=1, 2, ..., r)

[0096] To solve the above, if the least squares method is used, a singular matrix may be encountered and the solution may not be found. The recursive least squares method (RLS) is created to avoid the inversion of high-order matrices and can improve the accuracy of parameter identification. The iterative formula for estimating parameter values ​​is:

[0097]

[0098] Here you set the initial value P0=αI. (α is a large value, which can be 10 5 ).

[0099] 5. Model Testing

[0100] In order to verify the universality and generalization ability of the model, the established model was tested. The specific steps are the same as the training set, except that the test set data group is different. The test data is extracted from a section of the unit load increase stage, with a total of 800 sets of data. The distribution of relevant data indicators is shown in the following table:

[0101]

[0102] Under the test set, the identification results of the gas turbine TS fuzzy model are as follows: Figure 4 shown.

[0103] Under the test set, the output variable evaluation indicators of the gas turbine TS fuzzy model are shown in the following table:

[0104]

[0105] According to the model identification results and evaluation indicators of the test set, the model output fits well with the actual output, and the values ​​of the mean square error and the determination coefficient are also within an acceptable range. The tested TS fuzzy model has good model generalization ability.

[0106] 6. Dynamic Matrix Predictive Control Algorithm (DMC)

[0107] Dynamic Matrix Control (DynamiDynami, DMC) is one of the earliest predictive control representative algorithms in model predictive control that originated from industrial processes. It is a predictive control algorithm based on the system step response curve. It is widely used in industrial fields such as chemical industry and energy production. It combines linear programming and constraint control problems in process control, solves static optimal problems, and is suitable for predictive control of linear controlled objects with self-balancing. In recent years, the DMC algorithm is also used to deal with predictive control problems with nonlinear systems, such as controllers designed in combination with PID, self-disturbance rejection, etc., which have good decoupling and nonlinear processing capabilities and have achieved good control effects. According to the prior art, the DMC algorithm includes three parts: the amount prediction model, rolling optimization and feedback correction. Since it belongs to the prior art, the present invention believes that those skilled in the art should know its related formulas, principles and derivation processes.

[0108] 7. Transfer Function

[0109] Improved particle swarm algorithm, specifically:

[0110] The first step is to define the variables to be identified, set the optimization objective function and stop condition, randomly generate a population of several particles, and initialize the position and velocity of the particles.

[0111] In the second step, the numerical value initially identified by Matlab is used as a reference to constrain the interval of particle search for optimal solution and speed up the search for the optimal solution;

[0112] The third step is to calculate the fitness value of each particle and record the current historical optimal solution and global optimal solution of each particle;

[0113] The fourth step is to determine the distance between each particle and the global optimal particle according to the formula, and calculate the inertia weight and learning factor of the particle respectively;

[0114] Step 5: Update the particle's velocity and position and record them, overwrite the original historical optimal solution and the global optimal solution, bring them into the objective function to solve, and determine whether the stop condition is met. If so, proceed to the next step, otherwise execute the third step;

[0115] The sixth step is to output the globally optimal particle individual, that is, the solution of the identified parameters.

[0116] The improved particle swarm algorithm parameter settings used in this paper are: the inertia weight is 0.9, the two learning factors are both set to 2, considering that each particle can be used as the optimal solution, the number of particles and the number of iterations are set to the same and are both 100, and the optimal objective function is set to the sum of the squares of the difference between the model identification output and the actual output. According to the preliminary determination of the parameters of the transfer function to be identified, the number domain of the parameters to be identified is defined according to the boundary of [-5,5]. The identified transfer function is stimulated by the original input through the lsim function in Matlab, and then compared with the actual output processed with zero initial value, as shown below. Figure 5 As shown in , the curves have basically the same trend and the error is within the accuracy requirement. Therefore, the new transfer function obtained by the above method is as shown below and will be used as the PID control of the model.

[0117]

[0118] 8. Dynamic Matrix Predictive Control Based on TTT Fuzzy Model

[0119] The fuzzy model is the first step to realize nonlinear predictive control. The multivariable linear system with dimensional output can be decomposed into r multi-input single-output systems according to the number of outputs. For the multi-input single-output system, cl clusters have been obtained through fuzzy clustering, and the sub-model description shown in the formula is obtained. It is further converted into an incremental form to meet the requirement of the DMC algorithm for the control increment to be a variable. At the same time, the influence of the constants in each linear model is eliminated. The deformed fuzzy rule is expressed as:

[0120]

[0121] in, represents the input vector of the l-th multi-input single-output system, d l Its dimension, y l Its output, represents the parameter in the sub-model, represents the output parameter of the sub-model pair, the cluster center cl and the linear model parameter All of them have been determined through offline models, and each sub-model is classified according to its corresponding membership It can be integrated into the global model of the multi-input single-output system, and the expression is as follows:

[0122]

[0123] Specifically for the study of gas turbine output control, the two outputs of the gas turbine can be considered separately, and the power output identification model and the exhaust temperature output identification model are established in combination with the DMC algorithm. First, the step response vector model of the corresponding operating conditions is obtained through the established gas turbine TS fuzzy model, and then the optimal input is calculated in combination with the DMC algorithm to optimize the control of the gas turbine output. The focus here is on the dynamic performance characteristics of the gas turbine, and the DMC algorithm performs feedback correction through the identification error of the model. In summary, a schematic diagram of the structure of the gas turbine output predictive control is established. Figure 6 .

[0124] IX. Control system simulation under given value disturbance

[0125] The given value disturbance is to observe the regulation effect of the predictive controller by changing the target value of the output. Specifically, by changing the target value of the exhaust temperature, the predictive controller will adjust the main control inputs such as the fuel quantity and IGV opening according to the new target setting value, and act on the gas turbine TS fuzzy model. The actual output of the gas turbine is sent to the predictive controller, thus forming a closed-loop feedback link, which increases the stability of the system. The schematic diagram of the control system structure under the given value disturbance is shown in the figure below. Figure 7 shown.

[0126] For the tracking experiment of power given value, this paper selects the stable working condition of the unit 700MW. At this time, the corresponding power of the gas turbine is 227MW. The power is set to 237MW, 217MW, 228MW and 235MW respectively. Each change is executed after the power output is stable. By continuously setting the power constant value of one increase, one decrease and then two increases, the tracking response ability of the system under TS combined with DMC algorithm is tested. The parameter settings selected by DMC algorithm are control time domain 60 and control time domain 1. The results combined with PID control effect are as follows Figure 8 shown.

[0127] It can be seen from the figure that the control effect of TS combined with DMC algorithm is better than that of conventional PID control, with good response tracking ability and basically no steady-state error. The corresponding regulation quality is shown in the following table:

[0128]

[0129] It can be concluded from the table that, for changes in the power setting value, the DMC algorithm based on the TS fuzzy model can respond quickly to the power output, has a very small overshoot, and has a short adjustment time; while the conventional PID control effect responds slightly slower than the hybrid algorithm, has a large overshoot, is prone to jitter, and prolongs the adjustment time.

[0130] 10. Control system simulation under external disturbance

[0131] External disturbance is to add a step signal between the controller and the controlled object to observe the regulation effect of the predictive controller. The external disturbance can reflect the ability of the controlled object to handle disturbance. This section uses irrelevant variables such as speed as the step signal of the disturbance. When the exhaust temperature output is stable, two different step signals, one rising edge and one falling edge, are set to simulate the external disturbance to observe the regulation and recovery ability of the exhaust temperature under the action of the predictive controller. The schematic diagram of the control system structure under step signal disturbance is shown in the figure. Fig. 9 shown.

[0132] In the experiment, the exhaust temperature was increased from 622℃ to 632℃, and the response index under the exhaust temperature transfer function was tested. For the setting of the step signal change in the experiment, a step response of the rising edge and the falling edge was performed at the simulation time of 500s and 1000s respectively. The adjusted controller parameters were still maintained, and the DMC algorithm based on the TS fuzzy model and the conventional PID were observed to handle the disturbance under the same fuel quantity disturbance. The dynamic characteristics of the exhaust temperature under the step signal disturbance are as follows: Fig.10 shown.

[0133] As can be seen from the figure, the exhaust temperature will fluctuate due to the change of the step signal, but the control effect of the DMC algorithm based on the TS fuzzy model is obviously superior to that of conventional PID control. The former can quickly return to the set value of the exhaust temperature, while the latter takes a certain amount of time to recover.

[0134] The table below shows that under the disturbance of the same step signal, the DMC algorithm based on the TS fuzzy model has a smaller overshoot, showing a certain ability to suppress disturbances. In terms of various time performances, it is better than the recovery ability after conventional PID disturbances.

[0135]

[0136] Based on the above simulation results and experimental data, the DMC algorithm based on the TS fuzzy model has better control effects than the traditional PID in target value tracking and step signal disturbance experiments. The result values ​​of the regulation qualities such as overshoot, peak time, and adjustment time are also basically smaller than the result values ​​of PID. The DMC algorithm based on the TS fuzzy model has the advantages of fast regulation speed, good stability, and strong disturbance suppression ability.

[0137] For those skilled in the art, various other corresponding changes and modifications can be made according to the technical solutions and concepts described above, and all of these changes and modifications should fall within the protection scope of the claims of the present invention.

Claims

1. A gas turbine TS model optimization method based on a multi-model predictive control algorithm, characterized in that: The method comprises: S1 data preprocessing, using five adjacent sampling points to perform cubic polynomial fitting to preprocess the data and improve data quality; S2 establishes the feature vector, that is, constructs the feature vector combination of input and output variables; S3 identifies the antecedent structure of the feature vector; S4 performs subsequent parameter identification on the data obtained in step S3; S5 obtains the TS fuzzy model; S6 obtains a step response vector model of the corresponding working condition in the TS fuzzy model obtained in step S5; S7 uses the DMC algorithm to find the optimal input to optimize the gas turbine output.

2. The gas turbine TS model optimization method based on the multi-model predictive control algorithm according to claim 1 is characterized in that: In step S1, a five-point cubic smoothing formula is used for data preprocessing, wherein: Let the original data be x, and the filtered data be y. The specific filtering algorithm rules are expressed as: x0 is the current data point in the data sequence, x -2 , x -1 , x1, x2 are the two points on the front and back sides of the current data point; y0 is the current data point in the data sequence after filtering, y -2 ,y -1 , y1, y2 are the two points on the front and back sides of the current data point after filtering.

3. The gas turbine TS model optimization method based on the multi-model predictive control algorithm according to claim 1 is characterized in that: The feature vector combination in step S2 is: Among them, X1 represents the characteristic vector combination of power, k represents the kth sampling time, which can also be understood as the order in the discrete system, X2 represents the characteristic vector combination of exhaust temperature, u1 represents the input variable fuel amount, u2 represents the input variable inlet guide vane opening, y1 represents the output variable engine power, and y2 represents the output variable exhaust temperature.

4. The gas turbine TS model optimization method based on the multi-model predictive control algorithm according to claim 1 is characterized in that: The step S3 adopts an improved FCM clustering algorithm, specifically: in, γ is the regularization factor.

5. The gas turbine TS model optimization method based on the multi-model predictive control algorithm according to claim 1 is characterized in that: It also includes improved particle swarm algorithm, specifically: The first step is to define the variables to be identified, set the optimization objective function and stop condition, randomly generate a population of several particles, and initialize the position and velocity of the particles. In the second step, the numerical value initially identified by Matlab is used as a reference to constrain the interval of particle search for optimal solution and speed up the search for the optimal solution; The third step is to calculate the fitness value of each particle and record the current historical optimal solution and global optimal solution of each particle; The fourth step is to determine the distance between each particle and the global optimal particle according to the formula, and calculate the inertia weight and learning factor of the particle respectively; Step 5: Update the particle's velocity and position and record them, overwrite the original historical optimal solution and the global optimal solution, bring them into the objective function to solve, and determine whether the stop condition is met. If so, proceed to the next step, otherwise execute the third step; Step 6: Output the globally optimal particle individual, i.e. the solution of the identified parameters; The accurate transfer function of the gas turbine site is obtained by improving the particle swarm algorithm, where: The improved transfer formula is:

Images