Neural network control method for controlling the speed of an industrial gas turbine

Abstract

The present disclosure provides a neural network control method for controlling the speed of an industrial gas turbine, comprising: obtaining a training sample data set, the training sample data set comprising a plurality of speed data samples, each speed data sample comprising a speed command value, a speed feedback value, a speed error value and label information; inputting the speed command value, the speed feedback value and the speed error value into an initial neural network to output a fuel quantity, wherein the activation function of the hidden layer neurons of the initial neural network is obtained by dimension reduction improvement on an activation function based on a sequence of inverse hyperbolic sine functions; adjusting model parameters of the initial neural network according to the speed command value, the speed feedback value, the speed error value and the fuel quantity, wherein the model parameters comprise connection weights and the number of hidden layer neurons; and determining the initial neural network obtained when the connection weights and the number of hidden layer neurons both satisfy a preset convergence condition as the neural network for controlling the speed of the industrial gas turbine.

Description

Technical Field

[0001] This disclosure relates to the field of gas turbine control, and more specifically to a neural network control method for controlling the speed of an industrial gas turbine. Background Technology

[0002] Industrial gas turbines are characterized by high nonlinearity, strong coupling, and complex operating conditions. Therefore, their control systems must meet the requirements of high efficiency, high flexibility, and high reliability. Utilizing advanced control technologies to further improve the regulation rate and control accuracy of gas turbines has always been a development direction for industrial gas turbine control. Neural network technology, with its simple principles, ease of design, and excellent ability to approximate arbitrary nonlinearities, holds significant potential for engineering applications in the field of industrial gas turbines.

[0003] However, in the process of realizing the inventive concept disclosed herein, the inventors discovered at least the following problems in the related technology: Since the activation function of the hidden layer neurons of the neural network is generally complex and there are mutual constraints between polynomials, the computational load is high when using the neural network to control the speed of the industrial gas turbine, which reduces the efficiency of controlling the speed of the industrial gas turbine. Summary of the Invention

[0004] In view of the above problems, this disclosure provides a neural network control method and apparatus for controlling the speed of industrial gas turbines to improve the control efficiency of the speed of industrial gas turbines, in order to at least partially solve the above problems.

[0005] According to a first aspect of this disclosure, a neural network control method for controlling the speed of an industrial gas turbine is provided. The method includes: acquiring a training sample dataset, the training sample dataset including multiple speed data samples, each speed data sample including a speed command value, a speed feedback value, a speed error value, and label information; inputting the speed command value, the speed feedback value, and the speed error value from each speed data sample into an initial neural network, and outputting fuel quantity, wherein the initial neural network includes hidden layer neurons, and the activation function of the hidden layer neurons is obtained by dimensionality reduction improvement of an activation function constructed based on an inverse hyperbolic sine function sequence; adjusting the model parameters of the initial neural network according to the speed command value, the speed feedback value, the speed error value, and the fuel quantity, wherein the model parameters include connection weights and the number of hidden layer neurons; and determining the initial neural network obtained when both the connection weights and the number of hidden layer neurons satisfy a preset convergence condition as the neural network for controlling the speed of an industrial gas turbine.

[0006] According to embodiments of this disclosure, adjusting the model parameters of the initial neural network based on the aforementioned speed command value, speed feedback value, speed error value, and fuel quantity includes: determining the output value of the hidden layer neurons based on the aforementioned speed command, speed feedback value, and speed error; determining the input activation matrix of the hidden layer neurons based on the output value of the hidden layer neurons; determining the output vector matrix of the initial neural network based on the label value, wherein the label value includes the actual fuel quantity associated with the aforementioned fuel quantity; and calculating the connection weights of the initial neural network based on the input activation matrix and the output vector matrix using a pseudo-inverse calculation function.

[0007] According to an embodiment of this disclosure, adjusting the model parameters of the initial neural network based on the aforementioned speed command value, speed feedback value, speed error value, and fuel quantity further includes: obtaining a speed command value function, a speed feedback value function, and a speed error value function based on the aforementioned speed command value, speed feedback value, and speed error value, respectively; initializing the number of function terms in the aforementioned speed command value function, speed feedback value function, and speed error value function; and determining the number of hidden layer neurons based on the function term values ​​when the number of function terms is greater than a preset threshold.

[0008] According to an embodiment of this disclosure, the method further includes determining the mean square error between the fuel quantity and the tag information when the value of the function term is less than or equal to a preset threshold for the number of function terms; and determining the number of hidden layer neurons based on the mean square error.

[0009] According to an embodiment of this disclosure, determining the number of hidden layer neurons based on the mean squared error includes: if the mean squared error is less than or equal to the preset mean squared error threshold, continuing to train the initial neural network for T rounds until the mean squared error between the fuel quantity output by the initial neural network and the tag information is greater than the preset mean squared error threshold, wherein, before each round of training the initial neural network, an addition operation is performed on the number of function terms to obtain an updated number of function terms, where T is greater than or equal to 1; if the mean squared error is greater than the preset mean squared error threshold, determining the number of hidden layer neurons based on the updated number of function terms.

[0010] According to embodiments of this disclosure, the dimensionality reduction improvement of the activation function constructed based on the inverse hyperbolic sine function sequence includes: determining a first input sequence, a second input sequence, and a third input sequence to be input to the activation function based on the speed command value, the speed feedback value, the speed error value, and the inverse hyperbolic sine function sequence; using the product of the first input sequence and the second input sequence as a first variable and the third input sequence as a second variable; and using the first variable and the second input variable as input variables to the activation function to reduce the dimensionality of the input variables to the activation function.

[0011] According to embodiments of this disclosure, the method further includes: acquiring operating data of an industrial gas turbine through a controller, wherein the operating data is obtained by tracking control or robust control; and determining the training sample dataset based on the operating data.

[0012] Another aspect of this disclosure provides a neural network control device for controlling the speed of an industrial gas turbine, comprising: a first acquisition module for acquiring a training sample dataset, the training sample dataset including speed command values, speed feedback values, speed error values, and label information; a first input module for inputting the speed command value, speed feedback value, and speed error value from each speed data sample into an initial neural network and outputting fuel quantity, wherein the initial neural network includes hidden layer neurons, and the activation function of the hidden layer neurons is obtained by dimensionality reduction improvement of an activation function constructed based on an inverse hyperbolic sine function sequence; a first adjustment module for adjusting the model parameters of the initial neural network according to the speed command value, speed feedback value, speed error value, and fuel quantity, wherein the model parameters include connection weights and the number of hidden layer neurons; and a first determination module for using the initial neural network obtained when the connection weights and the number of hidden layer neurons both satisfy a preset convergence condition as the neural network for controlling the speed of the industrial gas turbine.

[0013] Another aspect of this disclosure provides an electronic device, comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the aforementioned neural network control method for controlling the speed of an industrial gas turbine.

[0014] Another aspect of this disclosure provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the aforementioned neural network control method for controlling the speed of an industrial gas turbine.

[0015] According to embodiments of this disclosure, the activation function of the hidden layer neurons in the initial neural network is obtained by dimensionality reduction improvement of the activation function constructed based on an inverse hyperbolic sine function sequence. Because the activation function is constructed based on an inverse hyperbolic sine function sequence, the calculation relationship of the activation function can be simplified, and the constraint relationship between polynomials in the function can be removed. Furthermore, dimensionality reduction improvement of the activation function can reduce the dimension of the variables input to the hidden layer activation function, simplifying the complexity of the hidden layer neurons. This can reduce the computational load required for the neural network to control the speed of an industrial gas turbine, thus at least partially overcoming the problem of low speed control efficiency caused by the high computational load of the neural network in related technologies, thereby improving the control efficiency of the speed of the industrial gas turbine. Attached Figure Description

[0016] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0017] Figure 1 This schematically illustrates a system architecture diagram of a neural network control method and apparatus for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure;

[0018] Figure 2 A flowchart illustrating a neural network control method for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure is shown schematically.

[0019] Figure 3 A schematic diagram illustrating the structure of a neural network for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure is shown.

[0020] Figure 4 A flowchart illustrating the determination of neural network model parameters according to an embodiment of the present disclosure is shown schematically.

[0021] Figure 5(a) schematically illustrates a numerical diagram of actual fuel quantities according to an embodiment of the present disclosure;

[0022] Figure 5(b) schematically illustrates a numerical diagram of the target fuel quantity obtained by a neural network for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure;

[0023] Figure 5(c) schematically illustrates a graph of the relative error between the actual fuel quantity and the target fuel quantity according to an embodiment of the present disclosure;

[0024] Figure 6 The diagram illustrates the effect of speed control using a neural network for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure.

[0025] Figure 7The schematic diagram illustrates a structural block diagram of a training apparatus for a neural network for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure;

[0026] Figure 8 A block diagram schematically illustrates an electronic device suitable for implementing a neural network control method for controlling the speed of an industrial gas turbine, according to an embodiment of the present disclosure. Detailed Implementation

[0027] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0028] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0029] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0030] When using expressions such as "at least one of A, B, and C," the expression should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.). Similarly, when using expressions such as "at least one of A, B, or C," the expression should generally be interpreted in accordance with the meaning commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, or C" should include, but is not limited to, systems having A alone, having B alone, having C alone, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.).

[0031] During the operation of industrial gas turbines, to meet wide-ranging power demand variations, the control system needs to rapidly and accurately regulate the turbine speed for power output matching. When the operating environment conditions of the industrial gas turbine change significantly, the control system also needs to adaptively adjust to maintain control performance and resist the influence of disturbances. Commonly used technologies include BP (back propagation) neural networks and RBF (radial basis function) neural networks. However, using these neural networks to control speed still requires cumbersome iterative calculations. For example, it may require trial-and-error methods to determine the optimal number of hidden layer neurons; or embedding fuzzy logic into the neural network, which further complicates the calculations and increases the computational load. BP and RBF neural networks have high computational resource requirements, and the control calculation time for complex systems is too long to meet the millisecond-level computation cycle requirements of industrial gas turbine control systems.

[0032] In view of this, this disclosure provides a neural network control method for controlling the speed of an industrial gas turbine, which reduces the computational complexity of the neural network, reduces the dependence on computing resources, and improves the control efficiency of the industrial gas turbine speed. Specifically, the method includes: acquiring a training sample dataset, which includes multiple speed data samples, each speed data sample including a speed command value, a speed feedback value, a speed error value, and label information; inputting the speed command value, speed feedback value, and speed error value from each speed data sample into an initial neural network, and outputting fuel quantity, wherein the initial neural network includes hidden layer neurons, and the activation function of the hidden layer neurons is obtained by dimensionality reduction improvement of the activation function constructed based on the inverse hyperbolic sine function sequence; adjusting the model parameters of the initial neural network according to the speed command value, speed feedback value, speed error value, and fuel quantity, wherein the model parameters include connection weights and the number of hidden layer neurons; and determining the initial neural network obtained when the connection weights and the number of hidden layer neurons both satisfy a preset convergence condition as the neural network for controlling the speed of the industrial gas turbine.

[0033] Figure 1 The diagram schematically illustrates a system architecture of a neural network control method and apparatus for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure.

[0034] like Figure 1 As shown, the system architecture 100 may include a neural network 101 for controlling the speed of an industrial gas turbine, an actuator 102, an industrial gas turbine 103, and a measuring device 104.

[0035] When controlling the speed of an industrial gas turbine, speed command values, speed feedback values, and speed error values ​​can be input into the neural network 101. Based on these data, the neural network 101 calculates the target fuel quantity required by the industrial gas turbine. The actuator 102 can measure the actual fuel in the industrial gas turbine 103 according to the target fuel quantity, ensuring that the amount of fuel entering the industrial gas turbine 103 for combustion matches the target fuel quantity. The measuring device 104 can measure the speed of the industrial gas turbine to obtain a speed feedback value, which can be input into the neural network 101. In this process, the speed of the gas turbine can be controlled by changing the fuel input.

[0036] The following will be based on Figure 1 The described system architecture, through Figures 2-6 A neural network control method for controlling the speed of an industrial gas turbine, according to a disclosed embodiment, will be described in detail.

[0037] Figure 2 A flowchart illustrating a neural network control method for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure is shown schematically.

[0038] like Figure 2 As shown, the neural network control method for controlling the speed of an industrial gas turbine in this embodiment includes operations S210 to S240.

[0039] During operation S210, a training sample dataset is obtained. The training sample dataset includes multiple speed data samples. Each speed data sample includes a speed command value, a speed feedback value, a speed error value, and label information.

[0040] In operation S220, the speed command value, speed feedback value, and speed error value in each speed data sample are input into the initial neural network, and the fuel quantity is output. The initial neural network includes hidden layer neurons, and the activation function of the hidden layer neurons is obtained by dimensionality reduction improvement of the activation function constructed based on the inverse hyperbolic sine function sequence.

[0041] During operation S230, the model parameters of the initial neural network are adjusted based on the speed command value, speed feedback value, speed error value, and fuel quantity. The model parameters include connection weights and the number of neurons in the hidden layer.

[0042] In operation S240, the initial neural network obtained when the connection weights and the number of hidden layer neurons both satisfy the preset convergence conditions is determined as the neural network for controlling the speed of industrial gas turbines.

[0043] According to embodiments of this disclosure, operating data of an industrial gas turbine can be collected by a controller, wherein the operating data is obtained by tracking control or robust control; and a training sample dataset is determined based on the operating data. Specifically, multiple speed data samples can be collected from the industrial gas turbine's operating data using a PID (Proportional-Integral-Derivative) controller, and the operating data is obtained by tracking control or robust control. The speed command value, speed feedback value, speed error value, label information, etc., included in this operating data can be used as a training sample dataset.

[0044] According to embodiments of this disclosure, the speed command value may include a preset target speed value, the speed feedback value may include the actual speed value of the industrial gas turbine, and the speed error value may be the difference between the speed command value and the speed feedback value.

[0045] Figure 3 A schematic diagram illustrates the structure of a neural network for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure.

[0046] like Figure 3 As shown, the neural network 300 for controlling the speed of an industrial gas turbine may include input layer neurons 301, hidden layer neurons 302, and output layer neurons 303. In this embodiment, the input layer neurons 301 may be set to three, which are used to input the speed command value, the speed feedback value, and the speed error value, respectively. Multiple hidden layer neurons 302 may be set, and one output layer neuron 303 may be set, used to output the fuel quantity.

[0047] According to embodiments of this disclosure, the inverse hyperbolic sine function is the inverse function of the hyperbolic sine function. To construct the activation functions of each neuron in the hidden layer of the neural network, three identical sequences of inverse hyperbolic sine functions can be used. Specifically, the inverse hyperbolic function sequence can be asinh(x), asinh(2x), asinh(3x), etc. Compared with activation functions constructed using general complex polynomials, such as those constructed using Laguerre polynomials or Chebyshev polynomials, the activation function constructed based on the inverse hyperbolic sine function sequence is simpler in form, simplifying the calculation relationship of the activation function. In one embodiment, there is a recursive relationship between the terms of the activation function polynomial constructed using general functions. The calculation formula of the later term is generally obtained from the calculation of the previous term or several terms, and the relationship of the later term needs to be constrained by the previous terms. However, the terms of the activation function polynomial constructed based on the inverse hyperbolic sine function sequence are independent of each other and have no recursive relationship, thus removing the constraint relationship between the terms in the polynomial.

[0048] According to embodiments of this disclosure, dimensionality reduction improvement can be understood as reducing the number of neurons in the hidden layer. Specifically, according to... Figure 3 The neural network structure shown in the diagram, after the speed command value, speed feedback value, and speed error value are input into the input layer, are then input into the hidden layer neurons. Based on the speed command value, speed feedback value, speed error value, and inverse hyperbolic sine function sequence, the first, second, and third input sequences to be input into the activation function can be determined. The product of the first and second input sequences is used as the first variable, and the third input sequence as the second variable. These first and second input variables are then used as input variables to the activation function, which also achieves dimensionality reduction of the input variables. Compared to conventional construction methods, this dimensionality reduction improvement method reduces the number of hidden layer neurons from the cube of the basic sequence to the square, significantly reducing the complexity of the hidden layer and further simplifying the mapping operations of the hidden layer neurons.

[0049] According to embodiments of this disclosure, connection weights and the number of hidden layer neurons are two parameters of the neural network. Connection weights can include the connection weights between hidden layer neurons and output layer neurons. During training, it is necessary to obtain the optimal connection weights and the optimal number of hidden layer neurons. In this case, the trained neural network can be considered to have been obtained. Preset convergence conditions can be used to determine whether the connection weights and the number of hidden layer neurons are optimal. For example, these conditions could be whether the mean square error between the output fuel value and the true label value meets an error threshold, or whether the number of function terms in the three input function sequences meets a function term count threshold. The three input function sequences are obtained based on the speed command value, speed feedback value, and speed error value.

[0050] According to embodiments of this disclosure, the activation function of the hidden layer neurons in the initial neural network is obtained by dimensionality reduction improvement of the activation function constructed based on an inverse hyperbolic sine function sequence. Because the activation function is constructed based on an inverse hyperbolic sine function sequence, the calculation relationship of the activation function can be simplified, and the constraint relationship between polynomials in the function can be removed. Furthermore, dimensionality reduction improvement of the activation function can reduce the dimension of the variables input to the hidden layer activation function, simplifying the complexity of the hidden layer neurons. This can reduce the computational load required for the neural network to control the speed of an industrial gas turbine, thus at least partially overcoming the problem of low speed control efficiency caused by the high computational load of the neural network in related technologies, thereby improving the control efficiency of the speed of the industrial gas turbine.

[0051] According to an embodiment of this disclosure, operation S230 may further include the following operations: determining the output value of the hidden layer neuron based on the rotation speed command, rotation speed feedback value, and rotation speed error; determining the input activation matrix of the hidden layer neuron based on the output value of the hidden layer neuron; determining the output vector matrix of the initial neural network based on the label value, wherein the label value includes the actual fuel quantity associated with the fuel quantity; and calculating the connection weights of the initial neural network based on the pseudo-inverse calculation function of the input activation matrix and the output vector matrix.

[0052] According to embodiments of this disclosure, compared to the gradient descent method used in general neural networks, embodiments of this disclosure use a direct weight determination method to obtain the optimal connection weights. The direct weight determination method can be understood as a method that obtains the connection weights between hidden layer neurons and output layer neurons based on the inputs of hidden layer neurons and the outputs of output layer neurons. The process of determining the optimal connection weights in embodiments of this disclosure can be as shown in formula (1).

[0053] w=pinv(Z)y (1)

[0054] Here, w can represent the optimal connection weight vector, and the determination process of w can be shown in formula (2). Z can represent the input activation matrix of the hidden layer neurons, and the determination process of Z can be shown in formula (3). pinv is the pseudo-inverse calculation function.

[0055]

[0056]

[0057] In formula (3), L in the matrix can represent the number of sample data collected. The output of the j-th hidden layer neuron can be the l-th real sample data input into the neural network. y is a vector composed of the output values ​​of the real sample data. The real sample data can be understood as label information or real fuel quantity. The determination process of y can be shown in formula (4). R L×H In this matrix, L represents the number of rows, and H represents the number of columns. R H×1 In this context, H represents the number of rows in the matrix.

[0058]

[0059] According to an embodiment of this disclosure, operation S230 may further include the following operations: obtaining the speed command value function, speed feedback value function, and speed error value function based on the speed command value, speed feedback value, and speed error value, respectively; initializing the number of function terms in the speed command value function, speed feedback value function, and speed error value function; and determining the number of hidden layer neurons based on the function term values ​​if the number of function terms is greater than a preset threshold.

[0060] According to an embodiment of this disclosure, when the value of the function term is less than or equal to a preset threshold for the number of function terms, the mean square error between the fuel quantity and the tag information is determined; based on the mean square error, the number of hidden layer neurons is determined. The process of obtaining the mean square error can be shown in formula (5).

[0061]

[0062] Here, HE can represent mean squared error, which can be a metric reflecting the difference between the output value of the neural network and the true label; L can represent the number of rows in the matrix, or it can be understood as the number of samples of rotational speed data; y l It can represent a real label, y m The mean squared error (MSE) can represent the output value of a neural network, reflecting how closely the output value approximates the true label value. The optimal connection weights are those where the MSE between the output value and the true label value is minimized.

[0063] According to an embodiment of this disclosure, determining the number of hidden layer neurons based on the mean squared error includes: if the mean squared error is less than or equal to a preset mean squared error threshold, continuing to train the initial neural network for T rounds until the mean squared error between the fuel quantity output by the initial neural network and the label information is greater than the preset mean squared error threshold, wherein, before each round of training the initial neural network, an addition operation is performed on the number of function terms to obtain an updated number of function terms, where T is greater than or equal to 1; if the mean squared error is greater than the preset mean squared error threshold, determining the number of hidden layer neurons based on the updated number of function terms.

[0064] Figure 4 A flowchart illustrating the determination of neural network model parameters according to an embodiment of the present disclosure is shown schematically.

[0065] Operate S401 to obtain speed data samples.

[0066] According to embodiments of this disclosure, rotational speed data samples (x l y l ), l = 1, 2, ..., L in x l It can represent rotational speed sample data, y l This can represent the output value vector of the neural network. Based on the speed command value, speed feedback value, and speed error value in the speed sample data, three function sequences can be obtained respectively. The number of function terms q in the three function sequences is initialized, for example, q = 2. In one embodiment, the three function sequences can be the same, only the input independent variables are different; the independent variables can be the speed command value, speed feedback value, and speed error.

[0067] Operation S402, determine HE≤HEmin or q≤q min Is +1 true?

[0068] According to embodiments of this disclosure, HE min It can represent the mean square error threshold, q min This can represent a threshold for the number of function terms. In one embodiment, the mean squared error threshold and the function term count threshold can be set in advance. For example, the function term count threshold can be the minimum number of terms, such as q. min =2. The mean squared error threshold can be set appropriately large because the process of determining the model parameters requires determining whether HE ≤ HE. min In HE≤HE min In this case, the model parameters can be further adjusted through the training process, where HE ≥ HE. min In this case, execution can be considered complete, as the model parameters have been determined. If the mean squared error threshold is small, it's possible that during the initial execution of operation S402, the error will directly occur because HE ≥ HE. min In such cases, subsequent operations cannot proceed, resulting in the inability to obtain optimal model parameters. Therefore, the mean squared error threshold should be set appropriately large. In this embodiment of the disclosure, HE... min Set to 10.

[0069] Operation S403 updates the input excitation matrix, determines the connection weights, and calculates the mean square error.

[0070] According to embodiments of this disclosure, when HE≤HE min or q≤q min If +1 is true, execute operation S403. In operation S403, the hidden layer input activation matrix Z, connection weight vector w, and mean square error HE can be calculated based on the currently obtained neural network structure.

[0071] Operation S404, determine if HE ≤ HE min Whether it is valid or not.

[0072] According to an embodiment of this disclosure, the mean square error obtained in operation S403 is used for another judgment. If the judgment is true, subsequent steps are executed. If the judgment is false, q is directly set to q+1, and operation S402 is executed again based on the updated function term threshold.

[0073] Operate S405 to execute HE min =HE,q min =q, and w opt =w.

[0074] According to an embodiment of this disclosure, in operation S404, HE≤HE minIf the condition is met, execute operation S405 and set q = q + 1. Then, execute operation S402 based on the updated mean squared error and the function term threshold.

[0075] Operate S406 to save the final calculation results.

[0076] According to an embodiment of this disclosure, in operation S402, HE≤HE min or q≤q min If none of the +1 conditions are met, execute operation S406. Save the minimum mean square error HE. min Optimal number of hidden layer neurons H opt =q min 3 and the optimal connection weight vector w opt The optimal neural network for controlling the speed of an industrial gas turbine can be obtained based on the optimal number of hidden layer neurons and the optimal connection weight vector.

[0077] According to embodiments of this disclosure, the activation function of the hidden layer neurons in the neural network is improved. Specifically, an inverse hyperbolic sine function sequence is used to improve the activation function, simplifying the mapping operation of the hidden layer and reducing the computational load. After improving the activation function construction using a dimensionality reduction improvement method, the number of hidden layer neurons is greatly reduced, further reducing the computational load. The above-mentioned improvement method can effectively determine the number of hidden layer neurons and avoid complex iterative solutions, allowing the optimal hidden layer-output layer connection weight vector to be obtained directly in one step. The neural network provided in this disclosure can also be used to control parameters for other tasks, as well as other types of gas turbine control.

[0078] According to embodiments of this disclosure, the method for controlling the speed of an industrial gas turbine using a neural network may further include the following operations: acquiring a speed command value, a speed feedback value, a speed error value, and an actual fuel quantity of the industrial gas turbine; inputting the speed command, speed feedback, and speed error into a neural network controller and outputting a target fuel quantity, wherein the neural network controller is capable of executing the aforementioned neural network control method for controlling the speed of the industrial gas turbine; and adjusting the speed feedback value based on the difference between the target fuel quantity and the actual fuel quantity.

[0079] According to embodiments of this disclosure, when controlling the gas turbine speed, normal tracking control or robust control operating data, such as speed command value, speed feedback value, speed error value, and actual fuel quantity, can be collected under a PID controller. The optimal connection weights and the optimal number of hidden layer neurons are trained using the aforementioned training method, thereby obtaining the optimal neural network controller structure. Using the speed command value, speed error value, and speed feedback value as input parameters, and the fuel quantity as the output parameter, this neural network is designed as a controller for controlling the gas turbine speed. The speed command, speed feedback, and speed error are input into this optimal neural network controller, which outputs the target fuel quantity, thereby controlling the actual speed based on the fuel quantity.

[0080] According to an embodiment of this disclosure, taking the step transition of a gas turbine speed from the same idle speed to the rated speed as an example, the specific process of the neural network controlling the gas turbine speed includes the following operations.

[0081] The operating data of the gas turbine under the PID controller, when the speed jumps from the synchronous idle speed to the rated speed, is collected as the speed data sample. The input values ​​include the speed command value, the speed feedback value, and the speed error value, and the output value is the fuel quantity.

[0082] Training neural networks can be done using methods such as direct weight determination and structure determination algorithms. The training results are as follows: Figures 5(a) to 5(c) As shown in Figure 5(a), a numerical diagram of the actual fuel quantity according to an embodiment of the present disclosure is shown; Figure 5(b) a numerical diagram of the target fuel quantity obtained by a neural network for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure is shown; Figure 5(c) a numerical diagram of the relative error between the actual fuel quantity and the target fuel quantity according to an embodiment of the present disclosure is shown.

[0083] Figures 5(a) to 5(c) The horizontal axis represents the number of speed data samples, and the vertical axis represents the relative error (%). Figure 5(a) shows the actual fuel quantity, Figure 5(b) shows the fuel quantity value fitted and output by the neural network, and Figure 5(c) shows the relative error between the actual and output fuel quantity values. The curve trends in Figure 5(a) and Figure 5(b) are consistent, indicating that the neural network fitting process is correct. The relative error in Figure 5(c) is mostly distributed near zero, indicating that the relative error is small, the neural network fitting process is relatively accurate, and the output fuel quantity value is closer to the actual value.

[0084] Figure 6 The diagram illustrates the effect of speed control using a neural network that controls the speed of an industrial gas turbine, according to an embodiment of the present disclosure.

[0085] like Figure 6As shown, after the neural network is trained, with the speed command value, speed feedback value, and speed error value as inputs and the fuel quantity as output, the trained neural network is determined as the speed controller of the gas turbine. This enables control from the synchronous idle speed to the rated speed in a stepwise manner, and the control effect is as follows: Figure 6 As shown. Figure 6 The horizontal axis represents time, and the vertical axis represents the normalized speed. The adjustment time from the idle speed to the rated speed is ts = 14.69s, and the steady-state error is ess = 0.001.

[0086] It should be noted that, unless it is explicitly stated that there is a sequential order of execution between different operations, or that there is a sequential order of execution between different operations in terms of technical implementation, the execution order between multiple operations may not be significant, and multiple operations may be executed simultaneously.

[0087] Based on the aforementioned neural network control method for controlling the speed of industrial gas turbines, this disclosure also provides a neural network control device for controlling the speed of industrial gas turbines. The following will be combined with... Figure 7 The device is described in detail.

[0088] Figure 7 A schematic block diagram of a neural network control device for controlling the speed of an industrial gas turbine according to an embodiment of the present disclosure is shown.

[0089] like Figure 7 As shown, the neural network control device 700 for controlling the speed of an industrial gas turbine in this embodiment includes a first acquisition module 710, a first input module 720, a first adjustment module 730, and a first determination module 740.

[0090] The first acquisition module 710 is used to acquire the training sample dataset, which includes speed command value, speed feedback value, speed error value and label information.

[0091] The first input module 720 inputs the speed command value, speed feedback value, and speed error value from each speed data sample into the initial neural network and outputs the fuel quantity. The initial neural network includes hidden layer neurons, and the activation function of the hidden layer neurons is obtained by dimensionality reduction improvement of the activation function constructed based on the inverse hyperbolic sine function sequence.

[0092] The first adjustment module 730 is used to adjust the model parameters of the initial neural network according to the speed command value, speed feedback value, speed error value and fuel quantity. The model parameters include connection weights and the number of hidden layer neurons.

[0093] The first determining module 740 is used to use the initial neural network obtained when the connection weights and the number of hidden layer neurons both satisfy the preset convergence conditions as the neural network for controlling the speed of industrial gas turbines.

[0094] According to embodiments of this disclosure, the first adjustment module further includes a first determining unit, a second determining unit, a third determining unit, and a calculation unit.

[0095] The first determining unit is used to determine the output value of the hidden layer neuron based on the rotation speed command, rotation speed feedback value, and rotation speed error.

[0096] The second determining unit is used to determine the input activation matrix of the hidden layer neurons based on the output values ​​of the hidden layer neurons.

[0097] The third determining unit is used to determine the output vector matrix of the initial neural network based on the label values, wherein the label values ​​include the actual fuel quantity associated with the fuel quantity.

[0098] The computation unit is used to calculate the input activation matrix and output vector matrix based on the pseudo-inverse computation function to obtain the connection weights of the initial neural network.

[0099] According to embodiments of this disclosure, the first adjustment module further includes a fourth determining unit, an initialization unit, and a fifth determining unit.

[0100] The fourth determining unit is used to obtain the speed command value function, speed feedback value function, and speed error value function based on the speed command value, speed feedback value, and speed error value, respectively.

[0101] The initialization unit is used to initialize the number of function terms for the speed command value function, speed feedback value function, and speed error value function.

[0102] The fifth determining unit is used to determine the number of hidden layer neurons based on the value of the function terms when the number of function terms is greater than a preset threshold.

[0103] According to embodiments of this disclosure, the neural network control device for controlling the speed of an industrial gas turbine further includes a second determining module and a third determining module.

[0104] The second determining module is used to determine the mean square error between fuel quantity and tag information when the value of the function term is less than or equal to a preset threshold for the number of function terms.

[0105] The third determination module is used to determine the number of hidden layer neurons based on the mean square error.

[0106] According to embodiments of this disclosure, the third determining module includes a training unit and a sixth determining unit.

[0107] The training unit is used to continue training the initial neural network for T rounds, provided that the mean squared error is less than or equal to a preset mean squared error threshold, until the mean squared error between the fuel quantity output by the initial neural network and the label information is greater than the preset mean squared error threshold. Before each round of training the initial neural network, an addition operation is performed on the number of function terms to obtain an updated number of function terms, where T is greater than or equal to 1.

[0108] The sixth determining unit is used to determine the number of hidden layer neurons based on the updated number of function terms when the mean square error is greater than a preset mean square error threshold.

[0109] According to embodiments of this disclosure, the first input module further includes a seventh determining unit, an eighth determining unit, and an input unit.

[0110] The seventh determining unit is used to determine the first input sequence, the second input sequence, and the third input sequence that need to be input to the activation function based on the speed command value, the speed feedback value, the speed error value, and the inverse hyperbolic sine function sequence.

[0111] The eighth determining unit is used to take the product of the first input sequence and the second input sequence as the first variable and the third input sequence as the second variable.

[0112] The input unit is used to take the first variable and the second input variable as input variables to the activation function, so as to reduce the dimensionality of the input variables to the activation function.

[0113] According to embodiments of this disclosure, the neural network control device for controlling the speed of an industrial gas turbine further includes a data acquisition module and a fourth determination module.

[0114] The data acquisition module is used to acquire operating data of the industrial gas turbine through the controller. The operating data is obtained by tracking control or robust control.

[0115] The fourth determination module is used to determine the training sample dataset based on the running data.

[0116] Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as hardware circuitry, such as Field Programmable Gate Arrays (FPGAs), Programmable Logic Arrays (PLAs), Systems-on-Chip, Systems-on-Substrate, Systems-on-Package, Application-Specific Integrated Circuits (ASICs), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0117] For example, any plurality of modules among the first acquisition module 710, the first input module 720, the first adjustment module 730, and the first determination module 740 may be combined into one module, or any one of these modules may be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules may be combined with at least part of the functionality of other modules and implemented in one module. According to embodiments of the present disclosure, at least one of the first acquisition module 710, the first input module 720, the first adjustment module 730, and the first determination module 740 may be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging the circuitry, or implemented in any one of the three methods of software, hardware, and firmware, or in a suitable combination of any of these. Alternatively, at least one of the first acquisition module 710, the first input module 720, the first adjustment module 730, and the first determination module 740 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0118] It should be noted that the neural network control device part for controlling the speed of industrial gas turbines in the embodiments of this disclosure corresponds to the neural network control method part for controlling the speed of industrial gas turbines in the embodiments of this disclosure. For a detailed description of the neural network control device part for controlling the speed of industrial gas turbines, please refer to the neural network control method part for controlling the speed of industrial gas turbines, which will not be repeated here.

[0119] Figure 8 A block diagram schematically illustrates an electronic device suitable for implementing a neural network control method for controlling the speed of an industrial gas turbine, according to an embodiment of the present disclosure.

[0120] like Figure 8 As shown, an electronic device 800 according to an embodiment of this disclosure includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage portion 808 into a random access memory (RAM) 803. The processor 801 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 801 may also include onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this disclosure.

[0121] RAM 803 stores various programs and data required for the operation of electronic device 800. Processor 801, ROM 802, and RAM 803 are interconnected via bus 804. Processor 801 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 802 and / or RAM 803. It should be noted that the programs may also be stored in one or more memories other than ROM 802 and RAM 803. Processor 801 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0122] According to embodiments of this disclosure, the electronic device 800 may further include an input / output (I / O) interface 805, which is also connected to a bus 804. The electronic device 800 may also include one or more of the following components connected to the I / O interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 810 as needed so that computer programs read from it can be installed into the storage section 808 as needed.

[0123] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0124] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 802 and / or RAM 803 and / or one or more memories other than ROM 802 and RAM 803 described above.

[0125] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0126] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined and / or combined in various ways, even if such combinations and / or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0127] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A neural network control method for controlling the speed of an industrial gas turbine, comprising: Obtain a training sample dataset, which includes multiple speed data samples. Each speed data sample includes a speed command value, a speed feedback value, a speed error value, and label information. The label information includes the actual fuel quantity associated with the fuel quantity. The speed command value, speed feedback value, and speed error value in each speed data sample are input into an initial neural network to output the fuel quantity. The initial neural network includes hidden layer neurons, and the activation function of the hidden layer neurons is obtained by dimensionality reduction improvement of the activation function constructed based on the inverse hyperbolic sine function sequence. Based on the speed command value, the speed feedback value, the speed error value, and the fuel quantity, the model parameters of the initial neural network are adjusted, wherein the model parameters include connection weights and the number of neurons in the hidden layer; The initial neural network obtained when both the connection weights and the number of hidden layer neurons satisfy a preset convergence condition is determined as the neural network for controlling the speed of an industrial gas turbine.

2. The method according to claim 1, wherein, Adjusting the model parameters of the initial neural network based on the speed command value, the speed feedback value, the speed error value, and the fuel quantity includes: The output value of the hidden layer neuron is determined based on the rotation speed command, the rotation speed feedback value, and the rotation speed error. The input activation matrix of the hidden layer neurons is determined based on the output values ​​of the hidden layer neurons; Based on the label information, determine the output vector matrix of the initial neural network; The connection weights of the initial neural network are obtained by calculating the input activation matrix and the output vector matrix based on the pseudo-inverse calculation function.

3. The method according to claim 1, wherein, Adjusting the model parameters of the initial neural network based on the speed command value, the speed feedback value, the speed error value, and the fuel quantity further includes: Based on the speed command value, the speed feedback value, and the speed error value, respectively, the speed command value function, the speed feedback value function, and the speed error value are obtained; Initialize the number of function terms in the speed command value function, the speed feedback value function, and the speed error value function; If the number of function terms is greater than a preset threshold for the number of function terms, the number of hidden layer neurons is determined based on the value of the function terms.

4. The method according to claim 3 further includes, when the value of the function term is less than or equal to a preset threshold for the number of function terms, Determine the mean square error between the fuel quantity and the tag information; The number of neurons in the hidden layer is determined based on the mean square error.

5. The method according to claim 4, wherein, The number of neurons in the hidden layer is determined based on the mean square error, including: If the mean square error is less than or equal to a preset mean square error threshold, the initial neural network is trained for T rounds until the mean square error between the fuel quantity output by the initial neural network and the tag information is greater than the preset mean square error threshold. Before each round of training the initial neural network, an addition operation is performed on the number of function terms to obtain an updated number of function terms, where T is greater than or equal to 1. If the mean squared error is greater than the preset mean squared error threshold, the number of hidden layer neurons is determined based on the updated number of function terms.

6. The method according to claim 1, wherein, The dimensionality reduction improvement of the activation function constructed based on the inverse hyperbolic sine function sequence includes: Based on the speed command value, the speed feedback value, the speed error value, and the inverse hyperbolic sine function sequence, determine the first input sequence, the second input sequence, and the third input sequence that need to be input to the activation function; The product of the first input sequence and the second input sequence is used as the first variable, and the third input sequence is used as the second variable; By using the first variable and the second variable as input variables to the activation function, the dimensionality of the input variables to the activation function is reduced.

7. The method according to claim 1, further comprising: The controller collects operating data from the industrial gas turbine, wherein the operating data is obtained by tracking control or robust control. The training sample dataset is determined based on the running data.

8. A neural network control device for controlling the speed of an industrial gas turbine, comprising: The first acquisition module is used to acquire a training sample dataset, which includes multiple speed data samples. Each speed data sample includes a speed command value, a speed feedback value, a speed error value, and label information. The label information includes the actual fuel quantity associated with the fuel quantity. The first input module inputs the speed command value, the speed feedback value, and the speed error value from each speed data sample into the initial neural network and outputs the fuel quantity. The initial neural network includes hidden layer neurons, and the activation function of the hidden layer neurons is obtained by dimensionality reduction improvement of the activation function constructed based on the inverse hyperbolic sine function sequence. The first adjustment module is used to adjust the model parameters of the initial neural network according to the speed command value, the speed feedback value, the speed error value and the fuel quantity, wherein the model parameters include connection weights and the number of neurons in the hidden layer; The first determining module is used to use the initial neural network obtained when the connection weights and the number of hidden layer neurons both satisfy a preset convergence condition as the neural network for controlling the speed of an industrial gas turbine.

9. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 7.

Images