An intelligent modeling method for heavy-duty gas turbines based on EffiCANet and KAN network

Abstract

The present invention discloses a heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network, belonging to the technical field of heavy-duty gas turbine intelligent modeling, comprising the following steps: S1, determining input variables and output variables; S2, collecting historical operating data of the heavy-duty gas turbine according to the results of S1, pre-processing the data, and then using variational mode decomposition to reduce noise; S3, constructing a heavy-duty gas turbine model based on EffiCANet and KAN network, initializing network parameters, inputting the data set processed in S2 into the model for training, and calculating the total loss; S4, based on the total loss, optimizing the EffiCANet network hyperparameters using an alpha evolutionary algorithm, stopping when the objective function value converges, obtaining the optimal hyperparameters, and completing model construction. The method protected by the present invention strengthens the interpretability and generalization ability of the model while ensuring that the model satisfies the mapping relationship between the actual operating data of the variables, and optimizes the network hyperparameters using an alpha evolutionary algorithm, thereby improving the accuracy of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of heavy-duty gas turbine intelligent modeling, and in particular to a heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network. Background Art

[0002] As the national energy development strategy increasingly focuses on energy and environmental issues, clean and renewable energy sources such as solar energy, wind energy, and tidal energy are gradually becoming a trend in the power generation field. Gas-steam combined cycle power generation is a clean and efficient thermal power generation method. It uses natural gas as fuel and generates electricity for users through the efficient and organic coordination between the gas turbine system and the steam turbine system. The power generation process emits relatively low levels of polluting gases such as sulfur oxides and nitrogen oxides, thus effectively promoting the realization of clean and low-carbon goals. As one of the key components of the combined cycle unit, the operational stability and efficiency of the heavy-duty gas turbine play a key role in the entire power generation process. Therefore, how to achieve the safe and stable operation of the heavy-duty gas turbine system is an urgent problem that needs to be solved.

[0003] In existing technology, control systems are often designed based on models that accurately reflect the process characteristics of the controlled object. An accurate controlled object model is a prerequisite for achieving efficient and stable operation of heavy-duty gas turbine systems. However, these systems are strongly coupled, nonlinear, and susceptible to external disturbances. Traditional data-driven modeling methods struggle to capture the system's internal physical laws, resulting in low model accuracy. To address this issue, the present invention provides an intelligent modeling method for heavy-duty gas turbines using an EffiCANet and KAN network. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent modeling method for a heavy-duty gas turbine based on EffiCANet and KAN network to solve the problems in the background technology.

[0005] To achieve the above object, the present invention provides a heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network, comprising the following steps:

[0006] S1. Based on the operating characteristics of heavy-duty gas turbines, the Spearman correlation coefficient is used to conduct correlation analysis on the variables to determine the input variables and output variables;

[0007] S2. Collect historical operating data of the heavy-duty gas turbine based on the results of S1, and use variational mode decomposition to reduce noise on the input variable data after preprocessing;

[0008] S3: Build a heavy-duty gas turbine model based on EffiCANet and KAN network, initialize network parameters, input the data set processed in S2 into the model for training, and calculate the total loss;

[0009] S4. Based on the total loss obtained in S3, the Alpha Evolution Algorithm is used to optimize the EffiCANet network hyperparameters. When the objective function value converges, the optimization is stopped to obtain the optimal hyperparameters of the network and complete the model construction.

[0010] Preferably, in S1, the Spearman correlation coefficient calculation formula is:

[0011]

[0012] Among them, R(x i )、R(y i ) are the values of two variables, are the mean values of variables x and y respectively, and n is the number of samples.

[0013] Preferably, in said S1, the input variables are the compressor inlet guide vane opening and the fuel quantity; and the output variables are the rotor speed, exhaust temperature, exhaust pressure, and output power.

[0014] Preferably, in S2, the pretreatment process is:

[0015] 1) Process missing values and outliers in the data;

[0016] 2) Perform maximum and minimum normalization on the state variables of different orders of magnitude in the collected data. The specific formula is:

[0017]

[0018] Among them, x * is the normalized data, x is the original data, and x min To process the minimum value of the original data of the characteristic variable, x max is the maximum value of the original data of the processed feature variable.

[0019] Preferably, in S2, the noise reduction process is as follows:

[0020] 1) Through variational optimization, the original signal is decomposed into k modal components with limited bandwidth and center frequency, while minimizing the sum of the bandwidths of the modal components. The constraint is that the original signal is equal to the sum of all modal components. The formula is expressed as:

[0021]

[0022] Where k is the number of center frequencies, j is the imaginary unit, ω k is the center frequency, u k is the modal component, δ(t) is the Dirac function, is the time derivative operator, t is the time variable, and f is the original signal to be denoised;

[0023] 2) The quadratic penalty parameter and Lagrange multiplier are introduced to transform the constraints into an unconstrained Lagrangian optimization problem, which can be expressed as:

[0024]

[0025] Where α is the quadratic penalty parameter, λ is the Lagrange multiplier, and f(t) is the original signal;

[0026] 3) By alternately updating the modal components, center frequencies, and Lagrange multipliers, the objective function is gradually optimized, and finally each modal component and the corresponding center frequency are obtained.

[0027] Preferably, the specific steps of S3 are as follows:

[0028] S31, input the data processed by S2 into the EffiCANet network in the form of matrix [B, M, H], where B represents the batch size, M represents the number of feature variables, and H represents the window length; each sliding window data x∈R M×H Increase the dimension x∈R by a non-squeezing operation M×1×H , the sequence is divided into N patches through a convolutional layer with a kernel size of P and a stride of S, and each input channel is mapped to D output channels, resulting in x emb ∈R M×D×N ;

[0029] in,

[0030] S32. Extract short-term and long-term features from multivariate time series through the time series large kernel decomposition convolution module, and use deep intelligent convolution to extract short-term dependencies, which can be expressed as:

[0031] x short =DWConv(x emb );

[0032] Deep dilated convolution is used to extract long-term dependencies, which can be expressed as:

[0033] x long =DWDConv(x emb );

[0034] Adding the two elements together gives:

[0035] x combined =x short +x long ;

[0036] S33. Model the dependency between variables through the inter-variable convolution module, fill the time dimension so that the time dimension can be divided by the window size, and implement convolution between multiple time windows, which is expressed as:

[0037]

[0038] Among them, N pad1 is the necessary padding length, T is the time dimension, and W is the window size;

[0039] Then, the head-tail filling strategy is adopted to obtain the dynamic pattern, which is expressed as:

[0040]

[0041] Among them, N left_pad2 The left padding length of the time dimension, N right_pad2 Pad the length on the right side of the time dimension; perform a one-dimensional group convolution on the channel dimension for each tensor, expressed as:

[0042] Y padded1 =Conv(X padded1 );

[0043] Y padded2 =Conv(X padded2 );

[0044] Among them, padded1 means standard padding, padded2 means head and tail padding, Conv means convolution, X padded1 is the input data after standard filling, X padded2 is the input data after padding at the beginning and end, Y padded1 is the data after standard padding and convolution, Y padded2 The data is padded at the head and tail and convolved;

[0045] The above results are combined and then convolved to refine the interaction between time and variables, and the convolution output is obtained, which is expressed as:

[0046] Y=Conv(Y padded1 +Y padded2 );

[0047] Among them, Y is the convolution output.

[0048] S34. Use the global time variable attention module to selectively enhance the feature learning of time dimension and variable dimension in multivariate time series. Given an input tensor Y∈R M×D×N , capturing global temporal dependencies and inter-variable dependencies, and combining temporal weights and variable weights with the convolution output as the output Y out ; Y out With input xemb Combined as input to the next module;

[0049] S35. Get the predicted value through the prediction layer Calculate predicted values With label value Y label The mean square error between them is used as the data loss term MSE in the total loss function. u ;

[0050] S36. Fitting the nonlinear equations inside the heavy-duty gas turbine system is expressed as:

[0051]

[0052] Obtain the derivative of the output of the EffiCANet network with respect to the input through automatic differentiation Using KAN network to obtain calculate and The mean square error between them is used as the physical loss term MSE in the total loss function f ;

[0053] S37. Calculate the total loss to guide the EffiCANet network parameter update. The total loss is expressed as:

[0054] MSE=MSE u +MSE f ,

[0055] Among them, MSE represents the total loss, MSE u Represents the data loss term; MSE f Represents the physical loss term.

[0056] Preferably, the specific steps of S34 are:

[0057] 1) Capture global temporal dependencies: Reshape the input tensor into Y temp ∈R (N×D)×M , apply global average pooling along the variable dimension, and then use a fully connected network to process the representation. The process is expressed as:

[0058] T pool =AvgPool(Y temp )∈R (N×D) ;

[0059] T atten =σ(W2·ReLU(W1·T pool ));

[0060] Among them, T atten is the time weight, AvgPool is the global average pool, T poolis the time-centered representation of the input tensor, W1 and W2 are weight matrices, σ is the sigmoid activation function, and ReLU is the rectified linear unit activation function;

[0061] 2) Capture inter-variable dependencies: Reshape the input tensor into Y var ∈R (M×D)×N , apply global average pooling along the time dimension, and then use a fully connected network to process the representation. The process is expressed as:

[0062] V pool =AvgPool(Y var )∈R (M×D) ;

[0063] V atten =σ(W4·ReLU(W3·V pool ));

[0064] Among them, V atten is the variable weight, V pool is the variable-centered representation of the input tensor, W3 and W4 are weight matrices;

[0065] 3) Combine the time and variable weights with the convolution output as the output, expressed as: Y out =σ(T atten ⊙V atten ⊙Y);

[0066] Among them, Y out represents the output, and ⊙ represents the Hadamard product.

[0067] 4) Further refine the feature representation and convert Y out with x emb Combined as input to the next module.

[0068] Preferably, the specific steps of S4 are:

[0069] S41. The initial hyperparameters of the EffiCANet network are regarded as candidate solution individuals, and the candidate solution individuals are initialized, which is expressed as:

[0070] X i =lb+(ub-lb)·rand(0,1,[1,D dim ]),i=1,2,…,N pop ;

[0071] Among them, X i represents the i-th candidate solution individual, lb and ub represent the lower and upper bounds of the hyperparameters respectively, rand represents the random number generator, D dim Indicates the dimension of candidate solutions, N pop represents the total number of candidate solutions;

[0072] S42. Determine the optimal candidate solution based on the fitness function, expressed as:

[0073] f(X i )=MSE u ′(X i );

[0074] in, It is a D dim dimensional vector; MSE u ′(X i ) represents the candidate solution X i The optimal data loss obtained by training the network;

[0075] S43. Update the value of the candidate solution, expressed as:

[0076]

[0077] in, E i represents the i-th evolution solution, t represents the current number of iterations, P represents the starting position of the evolution, β represents the attenuation factor, Δr i represents the i-th random step, θ represents the control parameter, W i and L i represents the sampling solution, f(W i )≤f(E i )≤f(L i );

[0078] S44. Apply boundary constraints to the candidate solutions. The specific formula is:

[0079]

[0080] Among them, E i,j is the i-th evolutionary solution of the j-th candidate solution individual;

[0081] S45. Determine whether the objective function value converges. If so, obtain the optimal hyperparameters of the EffiCANet network; otherwise, return to step S43.

[0082] Therefore, the present invention provides an intelligent modeling method for heavy-duty gas turbines based on EffiCANet and KAN networks. Based on the actual operating data of a heavy-duty gas turbine, the input and output variables of the neural network are selected through Spearman correlation analysis; variational mode decomposition is used to perform noise reduction on the data to reduce the model error at the source of the data; the EffiCANet network is used to extract the characteristics between the input data time and the variables, and a mapping relationship between the input and output variables is established, and the prediction error is used as a data loss term; the KAN network is used to fit the internal nonlinear equations of the heavy-duty gas turbine, and the prediction error is used as a physical loss term; the network is guided to update parameters based on data loss and physical loss; and the EffiCANet network hyperparameters are optimized through the alpha evolution algorithm to improve the accuracy of the model.

[0083] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 is a flow chart of a modeling method according to an embodiment of the present invention;

[0085] Figure 2 is a structural diagram of a modeling method according to an embodiment of the present invention;

[0086] Figure 3 This is a simplified structural diagram of a heavy-duty gas turbine system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0087] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0088] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0089] Example

[0090] like Figure 1-Figure 3 As shown, the present invention provides an intelligent modeling method for a heavy-duty gas turbine based on EffiCANet and KAN network, comprising the following steps:

[0091] S1. The heavy-duty gas turbine model primarily consists of three key components: the compressor, combustion chamber, and turbine. External air is compressed in stages by the compressor to form high-pressure gas. This gas then enters the combustion chamber, mixes with the injected fuel, and burns, producing high-temperature, high-pressure gas. This gas flows through the turbine, expands, and generates work, driving the turbine to rotate the generator at high speed, generating electricity. Taking full account of the operating characteristics of the heavy-duty gas turbine, the Spearman correlation coefficient was used to analyze the correlation between variables. The input variables were determined to be the compressor inlet guide vane opening and the fuel quantity; the output variables were determined to be the rotor speed, exhaust temperature, exhaust pressure, and output power.

[0092] The formula for calculating the Spearman correlation coefficient is:

[0093]

[0094] Among them, R(x i )、R(y i ) are the values of two variables, are the mean values of variables x and y respectively, and n is the number of samples.

[0095] S2. Based on the results of S1, historical operating data of heavy-duty gas turbines is collected, missing values and outliers of the data are preprocessed, and then the Min-Max normalization method is used to normalize the data set. The specific formula is:

[0096]

[0097] Among them, x * is the normalized data, x is the original data, and x min To process the minimum value of the original data of the characteristic variable, x max In order to process the maximum value of the original data of the characteristic variable, the variational mode decomposition is used to reduce the noise of the input variable data. The noise reduction process is as follows:

[0098] 1) Variational modal decomposition: Through variational optimization, the original signal is decomposed into k modal components with a limited bandwidth and a center frequency, while minimizing the sum of the bandwidths of the modal components and taking the sum of all modal components equal to the original signal as a constraint. The formula is expressed as:

[0099]

[0100] Where k is the number of center frequencies, j is the imaginary unit, ω k is the center frequency, u k is the modal component, δ(t) is the Dirac function, is the time derivative operator, t is the time variable, and f is the original signal to be denoised;

[0101] 2) To solve the constrained optimization problem, a quadratic penalty parameter and Lagrange multiplier are introduced to transform the constraints into an unconstrained Lagrangian optimization problem, which can be expressed as:

[0102]

[0103] Where α is the quadratic penalty parameter, λ is the Lagrange multiplier, and f(t) is the original signal;

[0104] 3) By alternately updating the modal components, center frequencies, and Lagrange multipliers, the objective function is gradually optimized, and finally each modal component and the corresponding center frequency are obtained.

[0105] S3. Build a heavy-duty gas turbine model based on the EffiCANet and KAN networks. The EffiCANet network is responsible for fitting the actual operating data, and the corresponding error loss is used as the data loss term of the total loss function. The KAN network is responsible for fitting the nonlinear equations within the system, and the corresponding error loss is used as the physical loss term of the total loss function. Initialize the network parameters, input the data set processed in S2 into the model for training, and calculate the total loss. The specific steps are as follows:

[0106] S31, input the data processed by S2 into the EffiCANet network in the form of matrix [B, M, H], where B represents the batch size, M represents the number of feature variables, and H represents the window length; each sliding window data x∈R M×H Increase the dimension x∈R by a non-squeezing operation M×1×H , ensuring that each variable can be processed independently. The sequence is divided into patches, mapping each input channel to D output channels, and obtaining x emb ∈R M×D×N ;

[0107] S32. Extract short-term and long-term features from multivariate time series through the time series large kernel decomposition convolution module, and use deep intelligent convolution to extract short-term dependencies, which can be expressed as:

[0108] x short =DWConv(x emb );

[0109] Deep dilated convolution is used to extract long-term dependencies, which can be expressed as:

[0110] x long =DWDConv(x emb );

[0111] Adding the two elements together gives:

[0112] x combined =xshort +x long ;

[0113] S33. Model the dependency between variables through the inter-variable convolution module, fill the time dimension so that the time dimension can be divided by the window size, and implement convolution between multiple time windows, which is expressed as:

[0114]

[0115] Among them, N pad1 is the necessary padding length, T is the time dimension, and W is the window size;

[0116] Then, we adopt the head-tail filling strategy to obtain various dynamic patterns, which can be expressed as:

[0117]

[0118] Among them, N left_pad2 The left padding length of the time dimension, N right_pad2 Fill the right side of the time dimension with length;

[0119] Perform a one-dimensional group convolution on each tensor in the channel dimension, expressed as:

[0120] X padded1 =Conv(X padded1 );

[0121] Y padded2 =Conv(X padded2 );

[0122] Among them, padded1 means standard padding, padded2 means head and tail padding, Conv means convolution, X padded1 is the input data after standard filling, X padded2 is the input data after padding at the beginning and end, Y padded1 is the data after standard padding and convolution, Y padded2 The data is padded at the head and tail and convolved;

[0123] The above results are combined and then convolved to further refine the interaction between time and variables, and the convolution output is obtained, which is expressed as:

[0124] Y=Conv(T padded1 +Y padded2 );

[0125] Among them, Y is the convolution output.

[0126] S34. Use the global time variable attention module to selectively enhance the feature learning of time dimension and variable dimension in multivariate time series. Given an input tensor Y∈R M×D×N, capturing global temporal dependencies and inter-variable dependencies, and combining temporal weights and variable weights with the convolution output as the output Y out ; Y out With input x emb Combined as the input of the next module; specifically:

[0127] 1) Capture global temporal dependencies: Reshape the input tensor into Y temp ∈R (N×D)×M , apply global average pooling along the variable dimension, and then use a fully connected network to process the representation. The process is expressed as:

[0128] T pool =AvgPool(Y temp )∈R (N×D) ;

[0129] T atten =σ(W2·ReLU(W1·T pool ));

[0130] Among them, T atten is the time weight, AvgPool is the global average pool, T pool is the time-centered representation of the input tensor, W1 and W2 are weight matrices, σ is the sigmoid activation function, and ReLU is the rectified linear unit activation function;

[0131] 2) Capture inter-variable dependencies: Reshape the input tensor into Y var ∈R (M×D)×N , apply global average pooling along the time dimension, and then use a fully connected network to process the representation. The process is expressed as:

[0132] V pool =AvgPool(Y var )∈R (M×D) ;

[0133] V atten =σ(W4·ReLU(W3·V pool ));

[0134] Among them, V atten is the variable weight, V pool is the variable-centered representation of the input tensor, W3 and W4 are weight matrices;

[0135] 3) Combine the time and variable weights with the convolution output as the output, expressed as: Y out =σ(T atten ⊙V atten ⊙Y);

[0136] Among them, Y outrepresents the output, and ⊙ represents the Hadamard product.

[0137] 4) Further refine the feature representation and convert Y out with x emb Combined as input to the next module.

[0138] S35. Get the predicted value through the prediction layer Calculate predicted values With label value Y label The mean square error between them is used as the data loss term MSE in the total loss function. u ;

[0139] S36. Fitting the nonlinear equations inside the heavy-duty gas turbine system is expressed as:

[0140]

[0141] Obtain the derivative of the output of the EffiCANet network with respect to the input through automatic differentiation Using KAN network to obtain calculate and The mean square error between them is used as the physical loss term MSE in the total loss function f ;

[0142] S37. Calculate the total loss to guide the EffiCANet network parameter update. The total loss is expressed as:

[0143] MSE=MSE u +MSE f .

[0144] S4. Based on the total loss obtained in S3, the Alpha Evolution Algorithm is used to optimize the EffiCANet network hyperparameters. When the objective function value converges, the algorithm stops and the optimal hyperparameters of the network are obtained, completing the model construction. The specific steps are as follows:

[0145] S41. The initial hyperparameters of the EffiCANet network are regarded as candidate solution individuals, and the candidate solution individuals are initialized, which is expressed as:

[0146] X i =lb+(ub-lb)·rand(0,1,[1,D dim ]),i=1,2,…,N pop ;

[0147] Among them, X i represents the i-th candidate solution individual, lb and ub represent the lower and upper bounds of the hyperparameters respectively, rand represents the random number generator, D dim Indicates the dimension of candidate solutions, Npop represents the total number of candidate solutions;

[0148] S42. Determine the optimal candidate solution based on the fitness function, expressed as:

[0149] f(X i )=MSE u ′(X i );

[0150] in, It is a D dim dimensional vector; MSE u ′(X i ) represents the candidate solution X i The optimal data loss obtained by training the network;

[0151] S43. Update the value of the candidate solution, expressed as:

[0152]

[0153] in, E i represents the θth evolution solution, t represents the current iteration number, P represents the starting position of the evolution, β represents the attenuation factor, Δr i represents the θth random step, θ represents the control parameter, W i and L i represents the sampling solution, f(W i )≤f(E i )≤f(L i );

[0154] S44. Apply boundary constraints to the candidate solutions. The specific formula is:

[0155]

[0156] Among them, E i,j is the i-th evolutionary solution of the j-th candidate solution individual;

[0157] S45. Determine whether the objective function value converges. If so, obtain the optimal hyperparameters of the EffiCANet network; otherwise, return to step S43.

[0158] Therefore, the present invention provides an intelligent modeling method for heavy-duty gas turbines based on EffiCANet and KAN network. The method adopts a structure that integrates EffiCANet and KAN network. EffiCANet is used to analyze actual operating data to obtain the mapping relationship between input and output variables. The KAN network is used to model the internal nonlinear equations of the heavy-duty gas turbine, and physical constraints are added to the established mapping relationship to improve the generalization ability of the model. The alpha evolution algorithm is used to optimize the network hyperparameters to improve the accuracy of the model.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network, characterized in that: The following steps are involved: S1. Based on the operating characteristics of heavy-duty gas turbines, the Spearman correlation coefficient is used to conduct correlation analysis on the variables to determine the input variables and output variables; S2. Collect historical operating data of the heavy-duty gas turbine based on the results of S1, and use variational mode decomposition to reduce noise on the input variable data after preprocessing; S3: Build a heavy-duty gas turbine model based on EffiCANet and KAN network, initialize network parameters, input the dataset processed in S2 into the model for training, and calculate the total loss. The specific steps are as follows: S31, the data processed by S2 is used as a matrix The form of is input into the EffiCANet network, where Indicates the batch size, represents the number of characteristic variables, Indicates the window length; each sliding window data Increase dimensionality through un-extrusion operations , the sequence is divided into patches, mapping each input channel to D output channels, and obtaining ; in, ; S32. Extract short-term and long-term features from multivariate time series through the time series large kernel decomposition convolution module, use deep intelligent convolution to extract short-term dependencies, use deep extended convolution to extract long-term dependencies, and add the two elements together; S33. Model the dependency between variables through the inter-variable convolution module, fill the time dimension so that the time dimension can be divided by the window size, and implement convolution between multiple time windows, which is expressed as: ； in, is the necessary filling length, is the time dimension, is the window size; Then, the head-tail filling strategy is adopted to obtain the dynamic pattern, which is expressed as: ； ； in, Fill the left side of the time dimension with length, Fill the right side of the time dimension with length; Perform a one-dimensional group convolution on each tensor in the channel dimension, expressed as: ； ； in, Indicates standard filling, Indicates padding at the beginning and end. is convolution, is the input data after standard filling, is the input data after padding at the beginning and end. is the data after standard padding and convolution, The data is padded at the head and tail and convolved; The results are combined and then convolved to obtain the convolution output, which is expressed as: ； in, is the convolution output; S34, using the global time variable attention module to selectively enhance the feature learning of the time dimension and variable dimension in multivariate time series, given an input tensor , capturing global temporal dependencies and inter-variable dependencies, and combining temporal weights and variable weights with convolutional outputs as output ;Will With input Combined as input to the next module; S35. Get the predicted value through the prediction layer , calculate the predicted value With label value The mean square error between them is used as the data loss term in the total loss function; S36. Fitting the nonlinear equations inside the heavy-duty gas turbine system is expressed as: ； Obtain the derivative of the output of the EffiCANet network with respect to the input through automatic differentiation , using KAN network to obtain ,calculate and The mean square error between them is used as the physical loss term in the total loss function; S37. Calculate the total loss to guide the EffiCANet network parameter update. The total loss is expressed as: ； in, represents the total loss, represents the data loss item; represents the physical loss term; S4. Based on the total loss obtained in S3, the Alpha Evolution Algorithm is used to optimize the EffiCANet network hyperparameters. When the objective function value converges, the optimization is stopped to obtain the optimal hyperparameters of the network and complete the model construction.

2. The heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network according to claim 1, characterized in that: In S1, the Spearman correlation coefficient calculation formula is: ； in, 、 are the values of two variables, 、 Variables and variables The mean value of , n is the number of samples.

3. The heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network according to claim 1, characterized in that: In S1, the input variables are the compressor inlet guide vane opening and the fuel quantity; the output variables are the rotor speed, exhaust temperature, exhaust pressure, and output power.

4. The heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network according to claim 1, characterized in that: In S2, the pre-processing process is: 1) Processing missing values and outliers in the data; 2) Perform maximum and minimum normalization on the state variables of different orders of magnitude in the collected data. The specific formula is: ； in, is the normalized data, is the original data, To process the minimum value of the original data of the characteristic variable, is the maximum value of the original data of the processed feature variable.

5. The heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network according to claim 1, characterized in that: In S2, the noise reduction process is as follows: 1) Through variational optimization, the original signal is decomposed into modal components with a finite bandwidth and a center frequency, while minimizing the sum of the bandwidths of the modal components. The constraint is that the original signal is equal to the sum of all modal components. The formula is expressed as: ； in, is the number of center frequencies, is the imaginary unit, is the center frequency, is the modal component, is the Dirac function, is the time derivative operator, is the time variable, is the original signal that needs noise reduction; 2) The quadratic penalty parameter and Lagrange multiplier are introduced to transform the constraints into an unconstrained Lagrangian optimization problem, which can be expressed as: ； in, is the quadratic penalty parameter, is the Lagrange multiplier, is the original signal; 3) By alternately updating the modal components, center frequencies, and Lagrange multipliers, the objective function is gradually optimized, and finally each modal component and the corresponding center frequency are obtained.

6. The heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network according to claim 1 is characterized in that The specific steps of S34 are: 1) Capturing global temporal dependencies: Reshape the input tensor into , apply global average pooling along the variable dimension, and then use a fully connected network to process the representation, which is expressed as: ； ； in, is the time weight, is the global average pooling, is the time-centric representation of the input tensor, and is the weight matrix, is the S-type activation function, is the rectified linear unit activation function; 2) Capturing inter-variable dependencies: Reshape the input tensor into , apply global average pooling along the time dimension, and then use a fully connected network to process the representation, which is expressed as: ； ； in, is the variable weight, is the variable-centric representation of the input tensor, and is the weight matrix; 3) Combine the time and variable weights with the convolution output as the output, expressed as: ； in, Indicates output, represents the Hadamard product; 4) Further refine the feature representation and and Combined as input to the next module.

7. The heavy-duty gas turbine intelligent modeling method based on EffiCANet and KAN network according to claim 1, characterized in that: The specific steps of S4 are: S41. The initial hyperparameters of the EffiCANet network are regarded as candidate solution individuals, and the candidate solution individuals are initialized, which is expressed as: ； in, Indicates the candidate solution individuals, and denote the lower and upper bounds of the hyperparameters, respectively. represents a random number generator, represents the dimension of the candidate solution, represents the total number of candidate solutions; S42. Determine the optimal candidate solution based on the fitness function, expressed as: ； in, is a dimensional vector; Represents a candidate solution based on The optimal data loss obtained by training the network; S43. Update the value of the candidate solution, expressed as: ； in, , Indicates the Second evolution and resolution, Indicates the current iteration number, represents the starting position of evolution, represents the attenuation factor, Indicates the A random step size, represents the control parameter, and represents the sampled solution, ; S44. Apply boundary constraints to the candidate solutions. The specific formula is: ； in, For the The first candidate solution individual Second evolution resolution; S45. Determine whether the objective function value converges. If so, obtain the optimal hyperparameters of the EffiCANet network; otherwise, return to step S43.

Images