CN115405428A - Neural network control method for controlling rotating speed of industrial gas turbine - Google Patents

Neural network control method for controlling rotating speed of industrial gas turbine Download PDF

Info

Publication number
CN115405428A
CN115405428A CN202211045286.6A CN202211045286A CN115405428A CN 115405428 A CN115405428 A CN 115405428A CN 202211045286 A CN202211045286 A CN 202211045286A CN 115405428 A CN115405428 A CN 115405428A
Authority
CN
China
Prior art keywords
rotating speed
value
neural network
function
hidden layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202211045286.6A
Other languages
Chinese (zh)
Inventor
郭森闯
刘月
肖波
刘培军
王子楠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Institute of Engineering Thermophysics of CAS
Original Assignee
Institute of Engineering Thermophysics of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Institute of Engineering Thermophysics of CAS filed Critical Institute of Engineering Thermophysics of CAS
Priority to CN202211045286.6A priority Critical patent/CN115405428A/en
Publication of CN115405428A publication Critical patent/CN115405428A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F02—COMBUSTION ENGINES; HOT-GAS OR COMBUSTION-PRODUCT ENGINE PLANTS
    • F02C—GAS-TURBINE PLANTS; AIR INTAKES FOR JET-PROPULSION PLANTS; CONTROLLING FUEL SUPPLY IN AIR-BREATHING JET-PROPULSION PLANTS
    • F02C9/00—Controlling gas-turbine plants; Controlling fuel supply in air- breathing jet-propulsion plants
    • F02C9/26—Control of fuel supply
    • F02C9/28—Regulating systems responsive to plant or ambient parameters, e.g. temperature, pressure, rotor speed
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Abstract

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

Description

Neural network control method for controlling rotating speed of industrial gas turbine
Technical Field
The present disclosure relates to the field of gas turbine control, and more particularly to a neural network control method for controlling the rotational speed of an industrial gas turbine.
Background
Industrial gas turbines have the characteristics of high nonlinearity, strong coupling and complex operating conditions, and therefore the control of industrial gas turbines needs to meet the requirements of high efficiency, high flexibility and high reliability. The use of advanced control techniques to further improve the turn-down rate and control accuracy of gas turbines has been a continuing trend in the control of industrial gas turbines. The neural network technology principle is simple, easy to design and good in any nonlinear approximation capability, so that the neural network has great engineering application potential in the field of industrial gas turbines.
However, the inventor finds that at least the following problems exist in the related art in the process of implementing the inventive concept of the present disclosure: because the relational expression of the activation function of the hidden layer neuron of the neural network is generally complex, and the polynomial have a mutual constraint relation, the calculation load is higher when the neural network is used for controlling the rotating speed of the industrial gas turbine, and the efficiency for controlling the rotating speed of the industrial gas turbine is reduced.
Disclosure of Invention
In view of the above problems, the present disclosure provides a neural network control method and apparatus for controlling a rotational speed of an industrial gas turbine, which improves efficiency of controlling the rotational speed of the industrial gas turbine, in order to at least partially solve the above problems.
According to a first aspect of the present disclosure, there is provided a neural network control method for controlling rotational speed of an industrial gas turbine, the method comprising: acquiring a training sample data set, wherein the training sample data set comprises a plurality of rotating speed data samples, and each rotating speed data sample comprises a rotating speed instruction value, a rotating speed feedback value, a rotating speed error value and label information; inputting the rotating speed instruction value, the rotating speed feedback value and the rotating speed error value in each rotating speed data sample into an initial neural network, and outputting fuel quantity, wherein the initial neural network comprises hidden layer neurons, and the activation function of the hidden layer neurons is obtained by performing dimension reduction improvement on an activation function constructed based on an inverse hyperbolic sine function sequence; adjusting model parameters of the initial neural network according to the rotating speed instruction value, the rotating speed feedback value, the rotating speed error value and the fuel quantity, wherein the model parameters comprise a connection weight and the number of the hidden layer neurons; and determining the initial neural network obtained when the connection weight and the number of the neurons of the hidden layer both meet the preset convergence condition as the neural network for controlling the rotating speed of the industrial gas turbine.
According to an embodiment of the present disclosure, adjusting the model parameter of the initial neural network according to the rotational speed command value, the rotational speed feedback value, the rotational speed error value, and the fuel amount includes: determining the output value of the hidden layer neuron according to the rotating speed instruction, the rotating speed feedback value and the rotating speed error; determining the input excitation matrix of the hidden layer neuron according to the output value of the hidden layer neuron; determining an output vector matrix of the initial neural network according to a tag value, wherein the tag value comprises a real fuel quantity associated with the fuel quantity; and calculating the input excitation matrix and the output vector matrix based on a pseudo-inverse calculation function to obtain the connection weight of the initial neural network.
According to an embodiment of the present disclosure, adjusting the model parameter of the initial neural network according to the rotational speed command value, the rotational speed feedback value, the rotational speed error value, and the fuel amount further includes: respectively obtaining a rotating speed instruction value function, a rotating speed feedback value function and a rotating speed error value function according to the rotating speed instruction value, the rotating speed feedback value and the rotating speed error value; initializing function terms of the rotating speed instruction value function, the rotating speed feedback value function and the rotating speed error value function; and under the condition that the number of the function terms is larger than a preset function term threshold value, determining the number of the neurons of the hidden layer according to the function term value.
According to an embodiment of the present disclosure, the method further includes, in a case that the function term value is less than or equal to a preset function term threshold value, determining a mean square error between the fuel quantity and the tag information; and determining the number of the hidden layer neurons according to the mean square error.
According to an embodiment of the present disclosure, determining the number of the hidden layer neurons according to the mean square error includes: under the condition that the mean square error is less than or equal to the preset mean square error threshold, continuing to train the initial neural network of T rounds until the mean square error between the fuel quantity output by the initial neural network and the label information is greater than the preset mean square error threshold, wherein before each round of training the initial neural network, adding operation is carried out on the function terms to obtain an updated function term, and T is greater than or equal to 1; and under the condition that the mean square error is larger than the preset mean square error threshold, determining the number of the neurons of the hidden layer according to the updated function term number.
According to the embodiment of the present disclosure, the performing the dimension reduction improvement on the activation function constructed based on the hyperbolic function sequence includes: determining a first input sequence, a second input sequence and a third input sequence which need to be input to the activation function according to the rotating speed instruction value, the rotating speed feedback value, the rotating speed error value and the hyperbolic function sequence; taking the product of the first input sequence and the second input sequence as a first variable, and taking the third input sequence as a second variable; and using the first variable and the second input variable as input variables input to the activation function to reduce the dimension of the input variables input to the activation function.
According to an embodiment of the present disclosure, the method further includes: collecting operating data of the industrial gas turbine by a controller, wherein the operating data is obtained by tracking control or robustness control; and determining the training sample data set according to the operating data.
Another aspect of the present disclosure also provides a neural network control apparatus for controlling a rotational speed of an industrial gas turbine, including: the device comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring a training sample data set, and the training sample data set comprises a rotating speed instruction value, a rotating speed feedback value, a rotating speed error value and label information; a first input module, configured to input the rotational speed instruction value, the rotational speed feedback value, and the rotational speed error value in each rotational speed data sample into an initial neural network, and output a fuel amount, where the initial neural network includes a hidden layer neuron, and an activation function of the hidden layer neuron is obtained by performing dimension reduction and improvement on an activation function constructed based on an inverse hyperbolic sine function sequence; a first adjusting module, configured to adjust a model parameter of the initial neural network according to the rotational speed instruction value, the rotational speed feedback value, the rotational speed error value, and the fuel amount, where the model parameter includes a connection weight and the number of hidden layer neurons; and the first determining module is used for taking the initial neural network obtained when the connection weight and the number of the neurons in the hidden layer both meet a preset convergence condition as the neural network for controlling the rotating speed of the industrial gas turbine.
Another aspect of the present disclosure also provides an electronic device, including: one or more processors; a storage device for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the above neural network control method for controlling rotational speed of an industrial gas turbine.
Yet another aspect of the present disclosure provides a computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to perform the above neural network control method for controlling rotational speed of an industrial gas turbine.
According to the embodiment of the disclosure, the activation function of the hidden layer neuron of the initial neural network is obtained by performing dimension reduction improvement on the activation function constructed based on the hyperbolic function sequence, because the activation function constructed based on the hyperbolic function sequence is used, the calculation relation of the activation function can be simplified, and the restriction relation between the polynomial and the polynomial in the function can be released; and then, the dimensionality reduction improvement is carried out on the activation function, the dimensionality of variables input to the hidden layer activation function can be reduced, and the complexity of the hidden layer neuron is simplified, so that the calculation load required by the neural network in the process of controlling the rotating speed of the industrial gas turbine can be reduced, the problem of low rotating speed control efficiency caused by high neural network calculation load in the related technology is at least partially solved, and the rotating speed control efficiency of the industrial gas turbine is further improved.
Drawings
The foregoing and other objects, features and advantages of the disclosure will be apparent from the following description of embodiments of the disclosure, which proceeds with reference to the accompanying drawings, in which:
FIG. 1 schematically illustrates a system architecture diagram of a neural network control method and apparatus for controlling the rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure;
FIG. 2 schematically illustrates a flow chart of a neural network control method for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure;
FIG. 3 schematically illustrates a block diagram of a neural network for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure;
FIG. 4 schematically illustrates a flow chart for determining neural network model parameters, in accordance with an embodiment of the present disclosure;
fig. 5 (a) schematically shows a numerical map of the actual fuel quantity according to an embodiment of the present disclosure;
FIG. 5 (b) is a graphical plot schematically illustrating a target fuel quantity derived by a neural network for controlling industrial gas turbine rotational speed, in accordance with an embodiment of the present disclosure;
fig. 5 (c) schematically shows a relative error value map between the actual fuel amount and the target fuel amount according to an embodiment of the present disclosure;
FIG. 6 schematically illustrates a graph of the effect of controlling rotational speed using a neural network for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure;
FIG. 7 schematically illustrates a block diagram of a training apparatus for a neural network for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure;
FIG. 8 schematically illustrates a block diagram of an electronic device suitable for implementing a neural network control method for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure.
Detailed Description
Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood that these descriptions are illustrative only and are not intended to limit the scope of the present disclosure. In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the disclosure. It may be evident, however, that one or more embodiments may be practiced without these specific details. Moreover, in the following description, descriptions of well-known structures and techniques are omitted so as to not unnecessarily obscure the concepts of the present disclosure.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The terms "comprises," "comprising," and the like, as used herein, specify the presence of stated features, steps, operations, and/or components, but do not preclude the presence or addition of one or more other features, steps, operations, or components.
All terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. It is noted that the terms used herein should be interpreted as having a meaning that is consistent with the context of this specification and should not be interpreted in an idealized or overly formal sense.
Where a convention analogous to "A, B and at least one of C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B and C" would include, but not be limited to, systems that have a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.). Where a convention analogous to "A, B or at least one of C, etc." is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B or C" would include, but not be limited to, systems that have a alone, B alone, C alone, a and B together, a and C together, B and C together, and/or A, B, C together, etc.).
In the operation process of the industrial gas turbine, in order to meet the requirement change of large-scale power, a control system is required to quickly and accurately regulate and control the rotating speed of the gas turbine so as to match the power output. When the operating environment condition of the industrial gas turbine is changed greatly, the control system also needs to be adjusted in a self-adaptive mode so as to maintain the control performance and resist the influence of interference factors. Generally used in the related art are a BP (back propagation) neural network and an RBF (Radial Basis Function) neural network. When the neural networks are used for controlling the rotating speed, a complex iterative calculation process is still needed, for example, a trial and error method is needed to determine the number of neurons in the optimal hidden layer; or fuzzy logic is embedded in the neural network, which makes the operation more complicated and more operation-burdened. The BP neural network and the RBF neural network have higher requirements on computing resources, and the control operation time of a complex system is too long, so that the requirement of an industrial gas turbine control system for realizing an operation period at a millisecond level cannot be met.
In view of this, the present disclosure provides a neural network control method for controlling the rotation speed of an industrial gas turbine, which reduces the computational complexity of the neural network, reduces the dependence on computational resources, and improves the control efficiency of the rotation speed of the industrial gas turbine. Specifically, the method comprises the following steps: acquiring a training sample data set, wherein the training sample data set comprises a plurality of rotating speed data samples, and each rotating speed data sample comprises a rotating speed instruction value, a rotating speed feedback value, a rotating speed error value and label information; inputting a rotating speed instruction value, a rotating speed feedback value and a rotating speed error value in each rotating speed data sample into an initial neural network, and outputting fuel quantity, wherein the initial neural network comprises hidden layer neurons, and an activation function of the hidden layer neurons is obtained by performing dimension reduction improvement on an activation function constructed based on an inverse hyperbolic sine function sequence; adjusting model parameters of the initial neural network according to the rotating speed instruction value, the rotating speed feedback value, the rotating speed error value and the fuel quantity, wherein the model parameters comprise a connection weight value and the number of hidden layer neurons; and determining the initial neural network obtained when the connection weight and the number of the neurons of the hidden layer both meet the preset convergence condition as the neural network for controlling the rotating speed of the industrial gas turbine.
FIG. 1 schematically illustrates a system architecture diagram of a neural network control method and apparatus for controlling the rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure.
As shown in FIG. 1, the system architecture 100 may include a neural network 101 for controlling the rotational speed of an industrial gas turbine, an actuator 102, an industrial gas turbine 103, and a measurement device 104.
When the rotation speed of the industrial gas turbine is controlled, a rotation speed command value, a rotation speed feedback value and a rotation speed error value can be input into the neural network 101, and the neural network 101 can obtain the target fuel quantity required by the industrial gas turbine according to the data. The actuator 102 can meter the actual fuel of the industrial gas turbine 103 according to the target fuel quantity so that the fuel quantity entering the industrial gas turbine 103 for combustion coincides with the target fuel quantity. The measuring device 104 may measure the rotational speed of the industrial gas turbine to obtain a rotational speed feedback value, which may be input into the neural network 101. During the above operation, the control of the gas turbine speed can be achieved by changing the fuel input.
The neural network control method for controlling the rotational speed of the industrial gas turbine according to the disclosed embodiment will be described in detail below with reference to fig. 2 to 6 based on the system architecture described in fig. 1.
FIG. 2 schematically illustrates a flow chart of a neural network control method for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure.
As shown in fig. 2, the neural network control method for controlling the rotational speed of the industrial gas turbine of the embodiment includes operations S210 to S240.
In operation S210, a training sample data set is obtained, where the training sample data set includes a plurality of rotation speed data samples, and each rotation speed data sample includes a rotation speed instruction value, a rotation speed feedback value, a rotation speed error value, and tag information.
In operation S220, the rotation speed instruction value, the rotation speed feedback value, and the rotation speed error value in each rotation speed data sample are input into an initial neural network, and the fuel quantity is output, where the initial neural network includes hidden layer neurons, and an activation function of the hidden layer neurons is obtained by performing dimension reduction and improvement on an activation function constructed based on an inverse hyperbolic sine function sequence.
In operation S230, a model parameter of the initial neural network is adjusted according to the rotation speed command value, the rotation speed feedback value, the rotation speed error value, and the fuel quantity, wherein the model parameter includes a connection weight and the number of hidden layer neurons.
In operation S240, an initial neural network obtained when both the connection weight and the number of hidden layer neurons satisfy a preset convergence condition is determined as a neural network for controlling the rotational speed of the industrial gas turbine.
According to an embodiment of the present disclosure, operational data of the industrial gas turbine may be collected by a controller, wherein the operational data is derived from tracking control or robust control; and determining a training sample data set according to the operating data. Specifically, the plurality of rotational speed data samples may be obtained by a PID (proportional-integral-derivative) controller to collect operation data of the industrial gas turbine, and the operation data is operation data controlled by tracking control or robustness. The rotating speed instruction value, the rotating speed feedback value, the rotating speed error value, the label information and the like included in the operating data can be used as a training sample data set.
According to an embodiment of the present disclosure, the rotation speed command value may include a preset target rotation speed value, the rotation speed feedback value may include an actual rotation speed value of the industrial gas turbine, and the rotation speed error value may be a difference value between the rotation speed command value and the rotation speed feedback value.
FIG. 3 schematically illustrates a block diagram of a neural network for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure.
As shown in FIG. 3, a neural network 300 for controlling the rotational speed of an industrial gas turbine may include an input layer neuron 301, an implicit layer neuron 302, and an output layer neuron 303. In the embodiment of the present disclosure, the input layer neurons 301 may be set to 3 for inputting the rotation speed instruction value, the rotation speed feedback value, and the rotation speed error value, respectively. The hidden layer neuron 302 may be set to a plurality, and the output layer neuron 303 may be set to 1, for outputting the fuel amount.
According to the embodiment of the disclosure, the inverse hyperbolic sine function is an inverse function of the hyperbolic sine function, and in order to construct the activation function of each neuron of the hidden layer of the neural network, three identical inverse hyperbolic sine function sequences can be utilized for construction. Specifically, the hyperbolic function sequence may be asinh (x), asinh (2 x), asinh (3 x), and the like. Compared with the activation function constructed by a general complex polynomial, for example, compared with the activation function constructed by a laguerre polynomial, a chebyshev polynomial and the like, the activation function constructed based on the hyperbolic sine function sequence has a simpler and more convenient hyperbolic sine function form, and the calculation relation of the activation function is simplified. In one embodiment, a recurrence relation exists among terms of an activation function polynomial of a general function structure, a calculation formula of a later term is obtained by calculation of a former term or former terms generally, and the relation formula of the later term needs to be restricted by the former term, and the terms of the activation function polynomial of the inverse hyperbolic sine function sequence structure are independent from each other and have no recurrence relation, so that the restriction relation among the terms in the polynomial is relieved.
According to the embodiment of the present disclosure, the dimension reduction improvement may be understood as reducing the number of hidden layer neurons, and specifically, according to the structure of the neural network shown in fig. 3, after a rotation speed instruction value, a rotation speed feedback value, and a rotation speed error value are input through the input layer, the input values are input into the hidden layer neurons. According to the rotating speed instruction value, the rotating speed feedback value, the rotating speed error value and the hyperbolic function sequence, a first input sequence, a second input sequence and a third input sequence which need to be input into the activation function can be determined. The product of the first input sequence and the second input sequence is used as a first variable, the third input sequence is used as a second variable, and the first variable and the second input variable are used as input variables input to the activation function, so that the dimension reduction of the input variables can be realized. Compared with a conventional construction method, the method has the advantages that the number of the neurons of the hidden layer can be reduced from the cubic power of the basic sequence to the quadratic power by adopting a dimension reduction improvement method, the complexity of the hidden layer is greatly reduced, and the mapping operation of the neurons of the hidden layer is further simplified.
According to an embodiment of the present disclosure, the connection weights and the number of hidden layer neurons are two parameters of the neural network, and the connection weights may include connection weights between the hidden layer neurons and the output layer neurons. In the training process, the optimal connection weight and the optimal number of hidden layer neurons need to be obtained, and in this case, a trained neural network can be considered to be obtained. The preset convergence condition may be used to determine whether the connection weight and the number of hidden layer neurons are the optimal conditions, for example, whether a mean square error between the output fuel value and the real tag value satisfies an error threshold, whether a number of function terms of the input three function sequences satisfies a function term threshold, and the like, where the input three function sequences are obtained based on the rotation speed command value, the rotation speed feedback value, and the rotation speed error value.
According to the embodiment of the disclosure, the activation function of the hidden layer neuron of the initial neural network is obtained by performing dimension reduction improvement on the activation function constructed based on the hyperbolic function sequence, because the activation function constructed based on the hyperbolic function sequence is used, the calculation relation of the activation function can be simplified, and the restriction relation between the polynomial and the polynomial in the function can be released; and then, the dimension reduction improvement is carried out on the activation function, the dimension of a variable input to the hidden layer activation function can be reduced, and the complexity of a hidden layer neuron is simplified, so that the calculation load required in the process of controlling the rotating speed of the industrial gas turbine by the neural network can be reduced, the problem of low rotating speed control efficiency caused by high calculation load of the neural network in the related technology is at least partially solved, and the rotating speed control efficiency of the industrial gas turbine is further improved.
According to an embodiment of the present disclosure, operation S230 may further include the operations of: determining the output value of the hidden layer neuron according to the rotating speed instruction, the rotating speed feedback value and the rotating speed error; determining an input excitation matrix of the hidden layer neuron according to the output value of the hidden layer neuron; determining an output vector matrix of an initial neural network according to a tag value, wherein the tag value comprises a real fuel quantity associated with the fuel quantity; and calculating the input excitation matrix and the output vector matrix based on a pseudo-inverse calculation function to obtain a connection weight of the initial neural network.
According to the embodiment of the disclosure, compared with a gradient descent method of a general neural network, the embodiment of the disclosure obtains the optimal connection weight by using a weight direct determination method. The direct weight determination method can be understood as a method that can obtain the connection weight between the hidden layer neuron and the output layer neuron according to the input of the hidden layer neuron and the output of the output layer neuron. The process of determining the optimal connection weight in the embodiment of the present disclosure may be as shown in formula (1).
w＝pinv(Z)y (1)
Wherein w may represent an optimal connection weight vector, and the determination process of w may be as shown in equation (2). Z may represent the input excitation matrix for the hidden layer neurons, and the process of determining Z may be as shown in equation (3). pinv is a pseudo-inverse calculation function.
Figure BDA0003822160560000101
Figure BDA0003822160560000102
Wherein, L in the matrix of formula (3) can represent the number of the collected sample data,
Figure BDA0003822160560000103
after the ith real sample data is input into the neural network, the output of the jth hidden layer neuron and y are vectors formed by output values of the real sample data, the real sample data can be understood as label information, and can also be a real fuel quantity, and the determination process of y can be shown as formula (4). R L×H Where L denotes the number of matrix rows and H denotes the number of matrix columns. R H×1 Where H denotes the number of matrix rows.
Figure BDA0003822160560000104
According to an embodiment of the present disclosure, operation S230 may further include the operations of: respectively obtaining a rotating speed instruction value function, a rotating speed feedback value function and a rotating speed error value function according to the rotating speed instruction value, the rotating speed feedback value and the rotating speed error value; initializing function terms of a rotating speed instruction value function, a rotating speed feedback value function and a rotating speed error value function; and under the condition that the number of the function terms is larger than a preset function term threshold value, determining the number of the neurons in the hidden layer according to the function term value.
According to the embodiment of the disclosure, under the condition that the function term value is less than or equal to the preset function term threshold value, determining the mean square error between the fuel quantity and the tag information; and determining the number of the hidden layer neurons according to the mean square error. The process of obtaining the mean square error can be shown in equation (5).
Figure BDA0003822160560000111
Wherein HE may represent the mean square error, which may be a metric reflecting the difference between the output value of the neural network and the true tag, L may represent the number of matrix lines, which may also be understood as the number of rotation speed data samples, y l Can represent a genuine tag, y m The output value of the neural network can be represented, and the degree of approximation of the output value of the neural network to the real tag value is reflected by the mean square error. And when the mean square error between the output value of the neural network and the real label value is minimum, the connection weight is the optimal connection weight.
According to an embodiment of the present disclosure, determining the number of hidden layer neurons according to mean square error comprises: under the condition that the mean square error is less than or equal to a preset mean square error threshold value, continuing to train the initial neural network of the T rounds until the mean square error between the fuel quantity output by the initial neural network and the label information is greater than the preset mean square error threshold value, wherein before each round of training of the initial neural network, adding operation is performed on the function terms to obtain an updated function term, and T is greater than or equal to 1; and under the condition that the mean square error is larger than a preset mean square error threshold value, determining the number of the neurons of the hidden layer according to the updated function term number.
Figure 4 schematically shows a flow chart for determining neural network model parameters according to an embodiment of the disclosure.
In operation S401, a rotation speed data sample is obtained.
According to an embodiment of the present disclosure, rotational speed data samples (x) l ，y l ) L =1,2, … …, x in L l Can represent the sample data of the rotation speed, y l A vector of output values of the neural network may be represented. And respectively obtaining three function sequences according to the rotating speed instruction value, the rotating speed feedback value and the rotating speed error value in the rotating speed sample data. The number of function terms q of the three function sequences is initialized, for example q is initialized to q =2. In one embodiment, the three function sequences may be the same, except that the input arguments are different, and the arguments may be a rotational speed command value, a rotational speed feedback value, and a rotational speed error.
Operation S402, determining HE is less than or equal to HE min Or q is not more than q min +1 is true.
According to an embodiment of the present disclosure, HE min Can represent the mean square error threshold, q min A function term threshold may be represented. In one embodiment, the mean square error threshold and the function term threshold may be set in advance. Illustratively, the function term number threshold may be a minimum of the term number, e.g., q min And (2). The setting of the mean square error threshold can be suitably large because the process of determining the model parameters requires a determination that HE ≦ HE min When HE is less than or equal to HE min Under the condition of (2), model parameters can be continuously adjusted through the training process, and the HE is more than or equal to HE min In this case, it can be understood that the parameters of the good model have been determined as ending execution. If the mean square error threshold is small, it may be possible at the beginning of operation S402, directly because HE ≧ HE min In this case, the subsequent operations cannot be performed continuously, and thus the optimal model parameters cannot be obtained, so the mean square error threshold should be set appropriately large. In the disclosed embodiment, HE min Set to 10.
Operation S403 updates the input excitation matrix, determines a connection weight, and calculates a mean square error.
According to embodiments of the present disclosure, HE ≦ HE min Or q is not more than q min If +1 is true, operation S403 is performed. In operation S403, a hidden layer input excitation matrix Z, a connection weight vector w, and a mean square error HE may be calculated according to the currently obtained neural network structure.
Operation S404, judging HE is less than or equal to HE min Whether or not this is true.
According to the embodiment of the present disclosure, the determination is performed again according to the mean square error obtained in operation S403. If yes, executing the subsequent steps. In the false condition, q = q +1 is directly made, and operation S402 is performed again based on the updated function term threshold.
Operation S405, performing HE min ＝HE，q min = q, and w opt ＝w。
According to an embodiment of the present disclosure, HE ≦ HE in operation S404 min If true, operation S405 is performed and q = q +1. And re-performs operation S402 based on the updated mean square error and the function term threshold.
In operation S406, the final calculation result is saved.
According to an embodiment of the present disclosure, HE ≦ HE in operation S402 min Or q is not more than q min If none of +1 is satisfied, operation S406 is performed. Preserving minimum mean square error HE min Optimum number of hidden layer neurons H opt ＝q min 3 And an optimal connection weight vector w opt . And obtaining the optimal neural network for controlling the rotating speed of the industrial gas turbine based on the optimal hidden layer neuron number and the optimal connection weight vector.
According to the embodiment of the disclosure, the activation function of the hidden layer neuron of the neural network is improved, specifically, the activation function is improved by adopting an inverse hyperbolic sine function sequence, so that the mapping operation of the hidden layer is simplified, and the operation load is reduced. After the activation function structure is improved by adopting a dimension reduction improvement method, the number of neurons in the hidden layer is greatly reduced, and the operation load is further reduced. The number of the neurons of the hidden layer can be effectively determined by the improved method, the complex iterative solving process is avoided, and the optimal weight vector of the connection between the hidden layer and the output layer can be directly obtained by one-step calculation. The neural network provided by the embodiment of the disclosure can also realize the control of other tasks, controlled parameters and other types of gas turbine control.
According to the embodiment of the disclosure, the method for controlling the rotating speed of the industrial gas turbine by using the neural network can further comprise the following operations: acquiring a rotating speed instruction value, a rotating speed feedback value, a rotating speed error value and an actual fuel quantity of the industrial gas turbine; inputting a rotating speed instruction, rotating speed feedback and a rotating speed error into a neural network controller, and outputting a target fuel quantity, wherein the neural network controller can execute the neural network control method for controlling the rotating speed of the industrial gas turbine; and adjusting the rotating speed feedback value according to the difference value between the target fuel quantity and the actual fuel quantity.
According to the embodiment of the disclosure, when controlling the rotation speed of the gas turbine, the operating data of normal tracking control or robust control, such as a rotation speed instruction value, a rotation speed feedback value, a rotation speed error value and an actual fuel quantity, can be collected under the PID controller. And training according to the training method to obtain the optimal connection weight and the optimal number of the neurons of the hidden layer, thereby obtaining the optimal structure of the neural network controller. And designing the neural network as a controller for controlling the rotating speed of the gas turbine by taking the rotating speed instruction value, the rotating speed error value and the rotating speed feedback value as input parameters and taking the fuel quantity as an output parameter. And inputting the rotating speed instruction, the rotating speed feedback and the rotating speed error into the optimal neural network controller, outputting a target fuel quantity, and further controlling the actual rotating speed according to the fuel quantity.
According to the embodiment of the disclosure, taking the rotation speed of the gas turbine stepped from the synchronous idle speed to the rated rotation speed as an example, the concrete process of controlling the rotation speed of the gas turbine by the neural network comprises the following operations.
Collecting operation data of a gas turbine, under a PID controller, of which the rotating speed is stepped from a synchronous idle speed to a rated rotating speed as a rotating speed data sample, wherein input values comprise a rotating speed instruction value, a rotating speed feedback value and a rotating speed error value, and an output value is fuel quantity.
The training using the neural network may use a weight direct determination method or a structure determination algorithm, and the training results are shown in fig. 5 (a) to 5 (c). Fig. 5 (a) schematically shows a numerical map of an actual fuel quantity according to an embodiment of the present disclosure; FIG. 5 (b) is a graphical plot schematically illustrating a target fuel quantity derived by a neural network for controlling industrial gas turbine rotational speed, in accordance with an embodiment of the present disclosure; fig. 5 (c) schematically shows a relative error value map between the actual fuel amount and the target fuel amount according to an embodiment of the present disclosure.
In fig. 5 (a) to 5 (c), the abscissa represents the number of rotation speed data samples, and the ordinate represents the relative error (%). Fig. 5 (a) is a real fuel amount value, fig. 5 (b) is a fuel amount value fitted and outputted by the neural network, and fig. 5 (c) is a relative error between the real fuel amount value and the outputted fuel amount value. The curve trend in fig. 5 (a) is consistent with the curve trend in fig. 5 (b), and it can be shown that the fitting process of the neural network is correct. Most of the relative errors in fig. 5 (c) are distributed near zero, which indicates that the relative errors are small, the fitting process of the neural network is accurate, and the output fuel value is closer to the true value.
FIG. 6 schematically illustrates a graph of the effect of controlling rotational speed using a neural network for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the disclosure.
As shown in fig. 6, after the training of the neural network is completed, the neural network obtained by training is determined as the rotational speed controller of the gas turbine by taking the rotational speed command value, the rotational speed feedback value and the rotational speed error value as inputs and taking the fuel amount as an output, so that the control from the synchronous idle speed step to the rated rotational speed can be realized, and the control effect can be shown in fig. 6. In fig. 6, the abscissa represents time, the ordinate represents normalized rotational speed, the adjustment time from the synchronous idle step to the rated rotational speed is ts =14.69s, and the steady-state error is ess =0.001.
It should be noted that, unless explicitly stated that there is an execution sequence between different operations or there is an execution sequence between different operations in technical implementation, the execution sequence between multiple operations may not be sequential, or multiple operations may be executed simultaneously in the flowchart in this disclosure.
Based on the neural network control method for controlling the rotating speed of the industrial gas turbine, the disclosure also provides a neural network control device for controlling the rotating speed of the industrial gas turbine. The apparatus will be described in detail below with reference to fig. 7.
FIG. 7 schematically illustrates a block diagram of a neural network control device for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure.
As shown in FIG. 7, the neural network control device 700 for controlling the rotational speed of an industrial gas turbine of this embodiment includes a first obtaining module 710, a first input module 720, a first adjusting module 730, and a first determining module 740.
The first obtaining module 710 is configured to obtain a training sample data set, where the training sample data set includes a rotation speed instruction value, a rotation speed feedback value, a rotation speed error value, and tag information.
The first input module 720 inputs the rotation speed instruction value, the rotation speed feedback value, and the rotation speed error value in each rotation speed data sample into an initial neural network, and outputs the fuel quantity, wherein the initial neural network includes hidden layer neurons, and an activation function of the hidden layer neurons is obtained by performing dimension reduction improvement on an activation function constructed based on an inverse hyperbolic sine function sequence.
The first adjusting module 730 is configured to adjust a model parameter of the initial neural network according to the rotational speed instruction value, the rotational speed feedback value, the rotational speed error value, and the fuel quantity, where the model parameter includes a connection weight and the number of hidden layer neurons.
And the first determining module 740 is configured to use the initial neural network obtained when the connection weight and the number of neurons in the hidden layer both meet the preset convergence condition as the neural network for controlling the rotation speed of the industrial gas turbine.
According to the embodiment of the disclosure, the first adjusting module further comprises a first determining unit, a second determining unit, a third determining unit and a calculating unit.
And the first determining unit is used for determining the output value of the hidden layer neuron according to the rotating speed instruction, the rotating speed feedback value and the rotating speed error.
A second determining unit for determining the input excitation matrix of the hidden layer neurons from the output values of the hidden layer neurons.
A third determination unit, configured to determine an output vector matrix of the initial neural network according to a tag value, where the tag value includes a true fuel quantity associated with the fuel quantity.
And the computing unit is used for computing the input excitation matrix and the output vector matrix based on a pseudo-inverse computing function to obtain the connection weight of the initial neural network.
According to an embodiment of the present disclosure, the first adjusting module further includes a fourth determining unit, an initializing unit, and a fifth determining unit.
And the fourth determining unit is used for respectively obtaining a rotating speed instruction value function, a rotating speed feedback value function and a rotating speed error value function according to the rotating speed instruction value, the rotating speed feedback value and the rotating speed error value.
And the initialization unit is used for initializing function terms of the rotating speed instruction value function, the rotating speed feedback value function and the rotating speed error value function.
And the fifth determining unit is used for determining the number of the neurons in the hidden layer according to the function term value under the condition that the function term number is greater than the preset function term number threshold value.
According to the embodiment of the disclosure, the neural network control device for controlling the rotating speed of the industrial gas turbine further comprises a second determination module and a third determination module.
And the second determination module is used for determining the mean square error between the fuel quantity and the label information under the condition that the function term value is less than or equal to the preset function term threshold value.
And the third determining module is used for determining the number of the hidden layer neurons according to the mean square error.
According to an embodiment of the disclosure, the third determination module includes a training unit, a sixth determination unit.
And the training unit is used for continuing training the initial neural network of the T round under the condition that the mean square error is less than or equal to a preset mean square error threshold value until the mean square error between the fuel quantity output by the initial neural network and the label information is greater than the preset mean square error threshold value, wherein before each round of training the initial neural network, the addition operation is performed on the function terms to obtain an updated function term, and T is greater than or equal to 1.
And the sixth determining unit is used for determining the number of the neurons in the hidden layer according to the updated function term number under the condition that the mean square error is larger than a preset mean square error threshold value.
According to an embodiment of the present disclosure, the first input module further includes a seventh determining unit, an eighth determining unit, and an input unit.
And the seventh determining unit is used for determining a first input sequence, a second input sequence and a third input sequence which need to be input into the activation function according to the rotating speed instruction value, the rotating speed feedback value, the rotating speed error value and the hyperbolic function sequence.
And the eighth determining unit is used for taking the product of the first input sequence and the second input sequence as the first variable and taking the third input sequence as the second variable.
And the input unit is used for taking the first variable and the second input variable as input variables input to the activation function so as to reduce the dimension of the input variables input to the activation function.
According to the embodiment of the disclosure, the neural network control device for controlling the rotating speed of the industrial gas turbine further comprises an acquisition module and a fourth determination module.
And the acquisition module is used for acquiring the operating data of the industrial gas turbine through the controller, wherein the operating data is obtained by tracking control or robustness control.
And the fourth determining module is used for determining a training sample data set according to the operating data.
Any number of modules, sub-modules, units, sub-units, or at least part of the functionality of any number thereof according to embodiments of the present disclosure may be implemented in one module. Any one or more of the modules, sub-modules, units, and sub-units according to the embodiments of the present disclosure may be implemented by being split into a plurality of modules. Any one or more of the modules, sub-modules, units, sub-units according to embodiments of the present disclosure may be implemented at least partially as a hardware circuit, 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 a package, an Application Specific Integrated Circuit (ASIC), or by any other reasonable means of hardware or firmware for integrating or packaging a circuit, or by any one of or a suitable combination of any of software, hardware, and firmware. Alternatively, one or more of the modules, sub-modules, units, sub-units according to embodiments of the disclosure may be at least partially implemented as a computer program module, which when executed may perform the corresponding functions.
For example, any plurality of the first obtaining module 710, the first inputting module 720, the first adjusting module 730, and the first determining module 740 may be combined and implemented in one module, or any one of the modules may be split into a plurality of 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 the other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the first obtaining module 710, the first inputting module 720, the first adjusting module 730, and the first determining module 740 may be at least partially implemented as a hardware circuit, 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 a package, an Application Specific Integrated Circuit (ASIC), or may be implemented by hardware or firmware in any other reasonable manner of integrating or packaging a circuit, or implemented by any one of three implementations of software, hardware, and firmware, or implemented by a suitable combination of any of them. Alternatively, at least one of the first obtaining module 710, the first inputting module 720, the first adjusting module 730 and the first determining module 740 may be at least partially implemented as a computer program module, which when executed may perform a corresponding function.
It should be noted that the portion of the neural network control device for controlling the rotation speed of the industrial gas turbine in the embodiment of the present disclosure corresponds to the portion of the neural network control method for controlling the rotation speed of the industrial gas turbine in the embodiment of the present disclosure, and the description of the portion of the neural network control device for controlling the rotation speed of the industrial gas turbine specifically refers to the portion of the neural network control method for controlling the rotation speed of the industrial gas turbine, and is not repeated herein.
FIG. 8 schematically illustrates a block diagram of an electronic device suitable for implementing a neural network control method for controlling rotational speed of an industrial gas turbine, in accordance with an embodiment of the present disclosure.
As shown in fig. 8, an electronic device 800 according to an embodiment of the present disclosure includes a processor 801 that 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 section 808 into a Random Access Memory (RAM) 803. The processor 801 may include, for example, a general purpose microprocessor (e.g., CPU), an instruction set processor and/or related chip sets and/or a special purpose microprocessor (e.g., application Specific Integrated Circuit (ASIC)), among others. The processor 801 may also include on-board memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flows according to embodiments of the present disclosure.
In the RAM 803, various programs and data necessary for the operation of the electronic apparatus 800 are stored. The processor 801, the ROM 802, and the RAM 803 are connected to each other by a bus 804. The processor 801 performs various operations of the method flows according to the embodiments of the present disclosure by executing programs in the ROM 802 and/or RAM 803. Note that the programs may also be stored in one or more memories other than the ROM 802 and the RAM 803. The processor 801 may also perform various operations of method flows according to embodiments of the present disclosure by executing programs stored in the one or more memories.
Electronic device 800 may also include input/output (I/O) interface 805, input/output (I/O) interface 805 also connected to bus 804, according to an embodiment of the present disclosure. Electronic device 800 may also include one or more of the following components connected to I/O interface 805: an input portion 806 including a keyboard, a mouse, and the like; an output section 807 including a signal such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker; a storage portion 808 including a hard disk and the like; and a communication section 809 including a network interface card such as a LAN card, a modem, or the like. 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 necessary. A removable medium 811 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 810 as necessary, so that a computer program read out therefrom is mounted on the storage section 808 as necessary.
The present disclosure also provides a computer-readable storage medium, which may be contained in the apparatus/device/system described in the above embodiments; or may exist separately and not be assembled into the device/apparatus/system. The computer-readable storage medium carries one or more programs which, when executed, implement the method according to an embodiment of the disclosure.
According to embodiments of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, which may include, for example but is not limited to: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. For example, according to embodiments of the present disclosure, a computer-readable storage medium may include one or more memories other than the ROM 802 and/or RAM 803 and/or ROM 802 and RAM 803 described above.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
It will be appreciated by a person skilled in the art that various combinations and/or combinations of the features recited in the various embodiments and/or claims of the present disclosure may be made, even if such combinations and/or combinations are not explicitly recited in the present disclosure. In particular, various combinations and/or combinations of the features recited in the various embodiments and/or claims of the present disclosure may be made without departing from the spirit or teaching of the present disclosure. All such combinations and/or associations are within the scope of the present disclosure.
The embodiments of the present disclosure have been described above. However, these examples are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the embodiments are described separately above, this does not mean that the measures in the embodiments cannot be used advantageously in combination. The scope of the disclosure is defined by the appended claims and equivalents thereof. Various alternatives and modifications can be devised by those skilled in the art without departing from the scope of the disclosure, and these alternatives and modifications are intended to fall within the scope of the disclosure.

Claims (10)

1. A neural network control method for controlling the rotational speed of an industrial gas turbine, comprising:
acquiring a training sample data set, wherein the training sample data set comprises a plurality of rotating speed data samples, and each rotating speed data sample comprises a rotating speed instruction value, a rotating speed feedback value, a rotating speed error value and label information;
inputting the rotating speed instruction value, the rotating speed feedback value and the rotating speed error value in each rotating speed data sample into an initial neural network, and outputting fuel quantity, wherein the initial neural network comprises hidden layer neurons, and the activation function of the hidden layer neurons is obtained by performing dimension reduction improvement on an activation function constructed based on an inverse hyperbolic sine function sequence;
adjusting model parameters of the initial neural network according to the rotating speed instruction value, the rotating speed feedback value, the rotating speed error value and the fuel quantity, wherein the model parameters comprise a connection weight value and the number of hidden layer neurons;
and determining the initial neural network obtained when the connection weight and the number of the neurons of the hidden layer both meet the preset convergence condition as the neural network for controlling the rotating speed of the industrial gas turbine.
2. The method of 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 comprises:
determining the output value of the hidden layer neuron according to the rotating speed instruction, the rotating speed feedback value and the rotating speed error;
determining an input excitation matrix of the hidden layer neuron according to the output value of the hidden layer neuron;
determining an output vector matrix of the initial neural network according to a tag value, wherein the tag value comprises a real fuel quantity associated with the fuel quantity;
and calculating the input excitation matrix and the output vector matrix based on a pseudo-inverse calculation function to obtain a connection weight of the initial neural network.
3. The method of 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 comprises:
respectively obtaining a rotating speed instruction value function, a rotating speed feedback value function and a rotating speed error value function according to the rotating speed instruction value, the rotating speed feedback value and the rotating speed error value;
initializing function terms of the rotating speed instruction value function, the rotating speed feedback value function and the rotating speed error value function;
and under the condition that the number of the function terms is larger than a preset function term threshold value, determining the number of the hidden layer neurons according to the function term value.
4. The method of claim 3, further comprising, in the event that the function term number value is less than or equal to a preset function term number threshold value,
determining a mean square error between the fuel quantity and the tag information;
and determining the number of the hidden layer neurons according to the mean square error.
5. The method of claim 4, wherein determining the number of hidden layer neurons from the mean square error comprises:
under the condition that the mean square error is less than or equal to the preset mean square error threshold, continuing to train the initial neural network of T rounds until the mean square error between the fuel quantity output by the initial neural network and the label information is greater than the preset mean square error threshold, wherein before each round of training of the initial neural network, adding operation is performed on the function terms to obtain an updated function term, and T is greater than or equal to 1;
and under the condition that the mean square error is larger than the preset mean square error threshold value, determining the number of the hidden layer neurons according to the updated function term number.
6. The method according to claim 1, wherein the dimensionality reduction improvement of the activation function constructed based on the sequence of inverse hyperbolic sinusoids comprises:
determining a first input sequence, a second input sequence and a third input sequence which need to be input into the activation function according to the rotating speed instruction value, the rotating speed feedback value, the rotating speed error value and the hyperbolic function sequence;
taking 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 taking the first variable and the second input variable as input variables input to the activation function so as to reduce the dimension of the input variables input to the activation function.
7. The method of claim 1, further comprising:
collecting, by a controller, operational data of an industrial gas turbine, wherein the operational data is derived from tracking control or robust control;
and determining the training sample data set according to the operating data.
8. A neural network control device for controlling the rotational speed of an industrial gas turbine, comprising:
the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring a training sample data set, and the training sample data set comprises a rotating speed instruction value, a rotating speed feedback value, a rotating speed error value and label information;
the first input module is used for inputting the rotating speed instruction value, the rotating speed feedback value and the rotating speed error value in each rotating speed data sample into an initial neural network and outputting fuel quantity, wherein the initial neural network comprises hidden layer neurons, and the activation function of the hidden layer neurons is obtained by performing dimension reduction improvement on an activation function constructed based on an inverse hyperbolic sine function sequence;
a first adjusting module, configured to adjust a model parameter of the initial neural network according to the rotational speed instruction value, the rotational speed feedback value, the rotational speed error value, and the fuel quantity, where the model parameter includes a connection weight and the number of hidden layer neurons;
and the first determining module is used for taking the initial neural network obtained when the connection weight and the number of the neurons in the hidden layer both meet a preset convergence condition as the neural network for controlling the rotating speed of the industrial gas turbine.
9. An electronic device, comprising:
one or more processors;
a storage device for storing one or more programs,
wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to perform the method recited in any of claims 1-7.
10. A computer readable storage medium having stored thereon executable instructions which, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 7.
CN202211045286.6A 2022-08-30 2022-08-30 Neural network control method for controlling rotating speed of industrial gas turbine Pending CN115405428A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211045286.6A CN115405428A (en) 2022-08-30 2022-08-30 Neural network control method for controlling rotating speed of industrial gas turbine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211045286.6A CN115405428A (en) 2022-08-30 2022-08-30 Neural network control method for controlling rotating speed of industrial gas turbine

Publications (1)

Publication Number Publication Date
CN115405428A true CN115405428A (en) 2022-11-29

Family

ID=84161651

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211045286.6A Pending CN115405428A (en) 2022-08-30 2022-08-30 Neural network control method for controlling rotating speed of industrial gas turbine

Country Status (1)

Country Link
CN (1) CN115405428A (en)

Similar Documents

Publication Publication Date Title
KR101807332B1 (en) Control device and method for calculating an output parameter for a controller
US8731737B2 (en) Microcontroller having a computing unit and a logic circuit, and method for carrying out computations by a microcontroller for a regulation or a control in a vehicle
US10466659B2 (en) Method for designing a non-linear controller for non-linear processes
CN110717600B (en) Sample pool construction method and device, and algorithm training method and device
CN102314533B (en) Methods and systems for matching a computed curve to a target curve
WO2021051556A1 (en) Deep learning weight updating method and system, and computer device and storage medium
CN105867138A (en) Stable platform control method and device based on PID controller
CN114026572A (en) Error compensation in analog neural networks
CN113659176B (en) Self-adaptive control method and device for hydrogen fuel cell
Li et al. Improved LSTM-based prediction method for highly variable workload and resources in clouds
US11506639B2 (en) Virtual sensing apparatus of structural vibration and operating method thereof
CN113537614A (en) Construction method, system, equipment and medium of power grid engineering cost prediction model
CN111624872B (en) PID controller parameter setting method and system based on adaptive dynamic programming
CN117242665A (en) Power system model calibration using measurement data
CN111856933B (en) Design method of self-adaptive controller of aircraft engine considering uncertainty
CN115405428A (en) Neural network control method for controlling rotating speed of industrial gas turbine
US11834066B2 (en) Vehicle control using neural network controller in combination with model-based controller
CN112926805A (en) Intelligent vertical well testing interpretation method and device based on deep reinforcement learning
He et al. Aerodynamic data fusion with a multi-fidelity surrogate modeling method
CN113343390B (en) Engine linearization modeling method based on neural network nonlinear model
CN104570757B (en) Method and device for operating an integrated control assembly
Grodzenskiy et al. Improving the Boundaries of Sequential Statistic Criteria by Computer Simulation
CN115009278B (en) Cruise control method, device, equipment and storage medium
CN110942478B (en) Texture integrity calculation method and calculation unit based on SystemC
CN117556725A (en) Flow field prediction method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination