Training method and use of compensation model for gas turbine

By training a multi-layer neural network compensation model, estimating the sensor parameter difference using the mechanism model, and processing different operation stages in segments, the commonality and accuracy of the gas turbine sensor parameter analysis balance is solved, and efficient analysis of the full operating conditions is achieved. Especially during the starting stage, stability is improved.

CN113761803BActive Publication Date: 2024-12-03INST OF ENGINEERING THERMOPHYSICS - CHINESE ACAD OF SCI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111067557.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-13
Publication Date
2024-12-03
Estimated Expiration
2041-09-13

AI Technical Summary

Technical Problem

The existing technology is difficult to design a sensor parameter analysis method with strong versatility, simple design method and high analytical accuracy. It is suitable for the full operating conditions of gas turbines, especially when the sensor parameters change drastically during the startup stage, resulting in the neural network model not being Stablize.

Method used

By obtaining the actual operating data of multiple sensor parameters, using the mechanism model to estimate the parameters to be parsed, calculating the difference as the sample label of the compensation model, training a multi-layer neural network compensation model, processing different operation stages in segments, and establishing suitable for full operation work analytical residual model of the situation.

Benefits of technology

The idea of ​​realizing the analysis of residual construction method is clear and simple, and is suitable for any gas turbine model, without considering complex factors such as variable heat. The model training is simple, the analytical accuracy is improved, and it is suitable for full operation conditions, making up for the gap in the startup stage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113761803B_ABST
    Figure CN113761803B_ABST
Patent Text Reader

Abstract

The present disclosure provides a method for training a compensation model of a gas turbine, comprising: obtaining actual operating data corresponding to a plurality of first sensor parameters and actual operating data corresponding to a plurality of second sensor parameters of the gas turbine; inputting the actual operating data corresponding to the first sensor parameters other than the parameter to be analyzed among the plurality of first sensor parameters into a mechanism model, and outputting an estimated value of the parameter to be analyzed; calculating the difference between the estimated value of the parameter to be analyzed and the actual operating data of the parameter to be analyzed, and obtaining the difference value of the parameter to be analyzed; using the actual operating data corresponding to the plurality of second sensor parameters as input values ​​of the compensation model, and using the difference value of the parameter to be analyzed as a sample label, and training the compensation model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the technical field of fault-tolerant control of gas turbines, and in particular to a training method and use of a compensation model of a gas turbine. Background Art

[0002] The gas turbine control system is an extremely important component of the entire gas turbine system. The control system has the functions of monitoring the working status of the gas turbine, controlling the start-up and speed regulation process of the gas turbine, etc. Its performance determines the variable operating performance, economy and safety of the gas turbine. With the development of gas turbine technology, the complexity of the control system continues to increase, and higher requirements are placed on the reliability of the control system. Therefore, the control system fault tolerance technology has become one of the important means to improve the quality and reliability of the control system. The control system fault tolerance includes fault tolerance for sensor faults and fault tolerance for actuator faults, among which sensor fault tolerance technology is the basis of the entire fault tolerance technology.

[0003] Gas turbines work for a long time under extremely harsh conditions such as high temperature, high pressure, and high stress. Sensors are prone to failure. Once an important parameter data is lost or the value is wrong, it will affect the accuracy and stability of the control system, causing the performance of the gas turbine to deteriorate or even shut down / stop. In actual engineering applications, redundant sensors are often configured to improve the reliability of measurement parameters, but this will increase the weight, volume, and development cost of the control system. Sensor resolution redundancy instead of hardware redundancy is the future direction of intelligent control system development.

[0004] There are two main ways to construct sensor analytical margin: mechanism analysis and neural network. The mechanism analysis method refers to the establishment of thermodynamic formulas between various parameters as analytical margins of certain parameters through in-depth analysis of the working principle of the gas turbine. However, this method often makes a lot of simplified assumptions on the real gas turbine, resulting in low analytical accuracy. If the accuracy is to be improved, the entire modeling process will be complicated. In recent years, with the development of artificial intelligence technology, machine learning algorithms represented by neural networks have been applied to sensor analytical margins. This method directly trains analytical models based on data without knowing the specific mathematical relationship between parameters, which simplifies modeling. However, during the entire operation of the gas turbine, some sensor parameters will change dramatically. For example, during the startup phase, after the combustion chamber is ignited, the compressor outlet temperature will rise sharply in a short time. This numerical change phenomenon is extremely unfavorable for neural network training, resulting in unstable results of the obtained neural network model, and a lot of effort is needed to improve the neural network. At present, there is no sensor parameter analytical margin design method with strong versatility, simple design method, high analytical accuracy, and suitable for the full operating conditions of gas turbines. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] Based on the above problems, the present disclosure provides a training method and use of a compensation model of a gas turbine to alleviate technical problems such as the complexity of the process of obtaining compensation parameters in the prior art.

[0007] (II) Technical solution

[0008] The present disclosure provides a method for training a compensation model of a gas turbine, comprising:

[0009] Acquire actual operation data respectively corresponding to a plurality of first sensor parameters and actual operation data respectively corresponding to a plurality of second sensor parameters of the gas turbine, wherein the plurality of first sensor parameters include the parameter to be analyzed, and the actual operation data of the first sensor parameters and the actual operation data of the second sensor parameters are collected at the same time;

[0010] Inputting actual operation data corresponding to the first sensor parameters other than the parameter to be analyzed among the plurality of first sensor parameters into a mechanism model, and outputting estimated values ​​of the parameter to be analyzed, wherein the mechanism model is used to characterize the functional relationship between the plurality of first sensor parameters;

[0011] Calculating the difference between the estimated value of the parameter to be analyzed and the actual operating data of the parameter to be analyzed to obtain the difference value of the parameter to be analyzed;

[0012] The actual operation data corresponding to the plurality of second sensor parameters are used as input values ​​of the compensation model, and the difference values ​​of the parameters to be analyzed are used as sample labels to train the compensation model.

[0013] In the embodiment of the present disclosure, the method further includes:

[0014] Dividing the operation process of the gas turbine into a plurality of operation stages;

[0015] Determine a target second sensor parameter corresponding to each of the operation stages;

[0016] For each of the operation stages, the actual operation data of the target second sensor parameter is used as an input value of the compensation model, and the difference value of the parameter to be analyzed is used as a sample label to train the compensation model.

[0017] In the embodiment of the present disclosure, the method further includes:

[0018] Determine a target compensation model corresponding to each of the operation stages;

[0019] Wherein, for each of the operation stages, using the actual operation data of the target second sensor parameter as the input value of the compensation model, taking the difference of the parameter to be analyzed as the sample label, training the compensation model includes:

[0020] For each of the operation stages, the actual operation data of the target second sensor parameter is used as an input value of the target compensation model, and the difference of the parameter to be analyzed is used as a sample label to train the target compensation model.

[0021] In an embodiment of the present disclosure, the plurality of first sensor parameters include at least one of the following:

[0022] The plurality of the second sensor parameters include at least one of the following:

[0023] Speed, ratio of IGV to fuel flow, compressor pressure ratio.

[0024] In the disclosed embodiment, the multiple operation phases include: a starting phase, a full-speed no-load phase, and a load adjustment phase.

[0025] In the embodiment of the present disclosure, the mechanism model is:

[0026]

[0027] Wherein, p1 is the compressor inlet pressure, unit: Pa;

[0028] p2 is the compressor outlet pressure, unit: Pa;

[0029] T1 is the compressor inlet temperature, unit: K;

[0030] T2 is the compressor outlet temperature, unit: K;

[0031] η c is the compressor efficiency;

[0032] k is the specific heat ratio of air.

[0033] In an embodiment of the present disclosure, the compensation model includes a multi-layer neural network, and the multi-layer neural network includes:

[0034] In the input layer, the number of neurons is consistent with the number of input variables;

[0035] The hidden layer is a standard neuron layer or a convolutional neural network layer, the number of the hidden layers is not less than one, and the number of neurons in each hidden layer is not less than the number of the input variables;

[0036] In the output layer, the number of neurons is consistent with the number of output variables.

[0037] In the embodiment of the present disclosure, the method further includes: performing standardization processing on the samples used for the compensation model training, thereby eliminating the influence of the numerical magnitude difference of the samples on the multi-layer neural network training process, and the standardization processing includes:

[0038] Obtaining the mean and standard deviation of the sample;

[0039] The sample is divided by the standard deviation after subtracting the mean.

[0040] In the embodiment of the present disclosure, training the multi-layer neural network further includes:

[0041] The stochastic gradient descent algorithm is used to gradually converge the cost function to the minimum value;

[0042] In the iterative process, the back propagation algorithm is used to calculate the gradient value of the cost function based on randomly selected small batch sample data.

[0043] The present disclosure also provides a compensation method using any one of the compensation models described above, comprising:

[0044] acquiring data of a second sensor parameter of the gas turbine to be measured;

[0045] Inputting data of the second sensor parameter of the gas turbine to be measured into the trained compensation model;

[0046] Outputting the compensation value of the gas turbine to be measured through the trained compensation model;

[0047] The resolution margin of the second sensor parameter is obtained by comparing the compensation value with the estimated value of the parameter to be resolved.

[0048] (III) Beneficial effects

[0049] It can be seen from the above technical solutions that the training method and use of the compensation model of the gas turbine disclosed in the present invention have at least one or part of the following beneficial effects:

[0050] (1)) The idea of ​​the redundancy construction method is clear and simple, and easy to implement. When establishing the mechanism model, there is no need to consider the influence of factors such as variable specific heat, compressor characteristics calculation, and working fluid composition changes. When establishing the compensation model, there is no drastic change in numerical values, and the model training is simple.

[0051] (2) This method is universal. The mechanism model in the analytical redundancy model is applicable to any gas turbine. When the gas turbine model is changed or the gas turbine is overhauled, only the compensation model needs to be retrained without major modifications to the entire analytical model.

[0052] (3) The analytical redundancy construction method adopts the idea of ​​segmentation. The entire analytical redundancy model is applicable to the full operating conditions of the gas turbine, filling the gap in the analytical redundancy of sensor parameters during the startup phase of the gas turbine. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The present invention is a flowchart of a method for training a compensation model of a gas turbine according to an embodiment of the present invention.

[0054] Figure 2 This is a schematic diagram of obtaining the analytical margin of the second sensor parameter of the compensation model of the gas turbine according to an embodiment of the present disclosure.

[0055] Figure 3 It is a schematic diagram of a compensation method of a compensation model of a gas turbine according to an embodiment of the present disclosure.

[0056] Figure 4 It is a comparison chart of the analytical results of the gas turbine outlet temperature of the compensation model of the gas turbine according to the embodiment of the present disclosure.

[0057] Figure 5 It is a schematic diagram of the analytical margin error of the second sensor parameter of the compensation model of the gas turbine according to the embodiment of the present disclosure. DETAILED DESCRIPTION

[0058] The present disclosure provides a training method and use of a compensation model of a gas turbine. When establishing a mechanism model of the method, it is not necessary to consider the influence of factors such as variable specific heat, compressor characteristic calculation, and working fluid component changes; when establishing the compensation model, there is no phenomenon of drastic numerical changes, and the model training is simple; the method is universal, and the analytical redundancy construction method adopts the idea of ​​segmentation. The entire analytical redundancy model is applicable to all operating conditions of the gas turbine, which can overcome the main shortcomings and deficiencies of the existing analytical redundancy construction.

[0059] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0060] In an embodiment of the present disclosure, a method for training a compensation model of a gas turbine is provided, such as Figure 1 As shown, the method comprises:

[0061] The actual operation data corresponding to a plurality of first sensor parameters and the actual operation data corresponding to a plurality of second sensor parameters of the gas turbine are obtained, wherein the plurality of first sensor parameters include a parameter to be analyzed, and the actual operation data of the first sensor parameter and the actual operation data of the second sensor parameter are collected at the same time; the actual operation data corresponding to the first sensor parameters other than the parameter to be analyzed among the plurality of first sensor parameters are input into a mechanism model, and an estimated value of the parameter to be analyzed is output, wherein the mechanism model is used to characterize the functional relationship between the plurality of first sensor parameters; the difference between the estimated value of the parameter to be analyzed and the actual operation data of the parameter to be analyzed is calculated to obtain the difference value of the parameter to be analyzed; the actual operation data corresponding to the plurality of second sensor parameters are used as input values ​​of a compensation model, and the difference value of the parameter to be analyzed is used as a sample label to train the compensation model.

[0062] In the disclosed embodiment, the sensor parameters include rotation speed, IGV (intake guide vane) angle and fuel flow rate.

[0063] In the embodiment of the present disclosure, a gas turbine is taken as an embodiment object, and the compressor outlet temperature T2 of the gas turbine is taken as an example. The above analytical redundancy construction method is used to construct the analytical redundancy model of T2.

[0064] Among them, the compressor component of the gas turbine can be regarded as an adiabatic compression device. The air is compressed by the compressor and becomes high-pressure air. According to the basic principles of thermodynamics, the relationship between the pressure and temperature of the air inlet and outlet in this process is: The mechanism model is:

[0065]

[0066] Wherein, p1 is the compressor inlet pressure, unit: Pa;

[0067] p2 is the compressor outlet pressure, unit: Pa;

[0068] T1 is the compressor inlet temperature, unit: K;

[0069] T2 is the compressor outlet temperature, unit: K;

[0070] η c is the compressor efficiency;

[0071] k is the specific heat ratio of air.

[0072] In the embodiment of the present disclosure, n c and k are unknown coefficients, which are 0.9 and 1.4 respectively.

[0073] The present disclosure also provides a method for training a compensation model of a gas turbine, comprising:

[0074] The untrained gas turbine control system compensation model is trained using samples in any of the training data sets mentioned above to obtain a trained gas turbine control system compensation model.

[0075] In the disclosed embodiment, the operating data of the gas turbine is collected, and the compressor inlet pressure, inlet temperature, and outlet pressure are input into the mechanism model to obtain an estimated value of T2, and the actual operating value of T2 is subtracted from the estimated value to obtain the error of the mechanism model.

[0076] The whole operation process of the gas turbine is divided into three stages: starting stage, full speed no-load stage and load adjustment stage. In the starting stage, the speed and the ratio of IGV to fuel flow are selected as the input of the compensation model; in the full speed no-load stage and the load adjustment stage, the ratio of IGV to fuel flow and the compressor pressure ratio are selected as the input of the compensation model.

[0077] The multiple operation stages include: starting stage, full speed no-load stage and load adjustment stage.

[0078] Through network architecture search, we obtain a neural network architecture with optimal performance, train compensation models at different stages, and finally obtain the parsing model, such as Figure 2 shown.

[0079] In the embodiment of the present disclosure, the compensation model of the gas turbine is a multi-layer neural network, which includes:

[0080] In the input layer, the number of neurons is consistent with the number of input variables;

[0081] The hidden layer is a standard neuron layer or a convolutional neural network layer. The number of hidden layers is not less than one, and the number of neurons in each hidden layer is not less than the number of input variables.

[0082] In the output layer, the number of neurons is consistent with the number of output variables.

[0083] In the embodiment of the present disclosure, the samples are standardized to eliminate the influence of the sample numerical magnitude difference on the multi-layer neural network training process. The standardization process includes:

[0084] Get the mean and standard deviation of the sample;

[0085] Subtract the sample mean and divide by the standard deviation.

[0086] In the embodiment of the present disclosure, training the multi-layer neural network further includes:

[0087] The stochastic gradient descent algorithm is used to gradually converge the cost function to the minimum value;

[0088] In the iterative process, the back propagation algorithm is used to calculate the gradient value of the cost function based on randomly selected small batch sample data.

[0089] The present invention also discloses a compensation method using any one of the compensation models described above, such as Figure 3 As shown, including:

[0090] acquiring data of a second sensor parameter of the gas turbine to be measured;

[0091] Inputting the data of the second sensor parameter of the gas turbine to be measured into the trained compensation model;

[0092] Outputting the compensation value of the gas turbine to be measured through the trained compensation model;

[0093] The resolution margin of the second sensor parameter is obtained by comparing the compensation value with the estimated value of the parameter to be resolved.

[0094] like Figure 4 The figure shows the comparison between the analytical margin model of the compressor outlet temperature and the actual operating data. The blue solid line represents the actual measured data of the compressor outlet temperature, the red dotted line represents the analytical data before compensation, that is, the data obtained by the mechanism model, and the yellow dotted line represents the analytical data after compensation. It can be seen that the analytical accuracy is significantly improved.

[0095] like Figure 5 As shown, it represents the relative error between the parameter analysis result after compensation and the actual data. It can be seen that the relative error is within a reasonable range.

[0096] So far, the embodiments of the present disclosure have been described in detail in conjunction with the accompanying drawings. It should be noted that the implementation methods not shown or described in the drawings or the body of the specification are all forms known to ordinary technicians in the relevant technical field and are not described in detail. In addition, the above definitions of each element and method are not limited to the various specific structures, shapes or methods mentioned in the embodiments, and ordinary technicians in the field can simply change or replace them.

[0097] Based on the above description, those skilled in the art should have a clear understanding of the compensation model training method and use of the gas turbine disclosed in the present invention.

[0098] In summary, the present invention provides a compensation model training method and use for a gas turbine. The idea of ​​constructing the method is clear, simple, and easy to implement. When establishing the mechanism model, there is no need to consider the influence of factors such as variable specific heat, compressor characteristic calculation, and working fluid component changes. When the compensation model is established, there is no drastic change in numerical value, and the model training is simple. The method is universal, and the mechanism model in the analytical redundancy model is applicable to any gas turbine. When the gas turbine model is replaced or the gas turbine is overhauled, it is only necessary to retrain the compensation model without making major modifications to the entire analytical model. The analytical redundancy construction method adopts the idea of ​​segmentation, and the entire analytical redundancy model is applicable to the full operating conditions of the gas turbine, which fills the gap in the analytical redundancy of sensor parameters during the startup phase of the gas turbine.

[0099] It should also be noted that the directional terms mentioned in the embodiments, such as "upper", "lower", "front", "back", "left", "right", etc., are only reference directions of the drawings and are not intended to limit the scope of protection of the present disclosure. Throughout the drawings, the same elements are represented by the same or similar reference numerals. Conventional structures or configurations will be omitted when they may cause confusion in the understanding of the present disclosure.

[0100] Moreover, the shapes and sizes of the components in the figures do not reflect the real size and proportion, but only illustrate the contents of the embodiments of the present disclosure. In addition, in the claims, any reference symbols between brackets shall not be constructed as limiting the claims.

[0101] Unless otherwise indicated, the numerical parameters in this specification and the appended claims are approximate values ​​and can vary according to the desired properties obtained through the content of the present disclosure. Specifically, all numbers used in the specification and claims to express the content of the composition, reaction conditions, etc., should be understood to be modified by the term "about" in all cases. In general, the meaning of the expression is to include a variation of ±10% in some embodiments, ±5% in some embodiments, ±1% in some embodiments, and ±0.5% in some embodiments by a specific number.

[0102] Furthermore, the word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements.

[0103] The ordinal numbers used in the specification and claims, such as "first", "second", "third", etc., to modify the corresponding elements, do not themselves mean that the elements have any ordinal numbers, nor do they represent the order of one element and another element, or the order in the manufacturing method. The use of these ordinal numbers is only used to clearly distinguish a component with a certain name from another component with the same name.

[0104] In addition, unless the steps are specifically described or must occur in sequence, the order of the above steps is not limited to the above list, and can be changed or rearranged according to the required design. And the above embodiments can be mixed and matched with each other or with other embodiments based on design and reliability considerations, that is, the technical features in different embodiments can be freely combined to form more embodiments.

[0105] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and set in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into multiple submodules or subunits or subcomponents. All features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination, except that at least some of such features and / or processes or units are mutually exclusive. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature that provides the same, equivalent or similar purpose. Furthermore, in a unit claim that lists several devices, several of these devices may be embodied by the same hardware item.

[0106] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various disclosed aspects, in the above description of the exemplary embodiments of the present disclosure, the various features of the present disclosure are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the following intention: the claimed disclosure requires more features than the features explicitly recited in each claim. More specifically, as reflected in the claims below, the disclosed aspects are less than all the features of the single embodiment disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present disclosure.

[0107] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present disclosure. It should be understood that the above description is only a specific embodiment of the present disclosure and is not intended to limit the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A method for training a compensation model of a gas turbine, comprising: Acquire actual operation data respectively corresponding to a plurality of first sensor parameters and actual operation data respectively corresponding to a plurality of second sensor parameters of the gas turbine, wherein the plurality of first sensor parameters include the parameter to be analyzed, and the actual operation data of the first sensor parameters and the actual operation data of the second sensor parameters are collected at the same time; Inputting actual operation data corresponding to the first sensor parameters other than the parameter to be analyzed among the plurality of first sensor parameters into the mechanism model, and outputting estimated values ​​of the parameter to be analyzed, wherein the mechanism model is used to characterize the functional relationship between the plurality of first sensor parameters, the first sensor parameters other than the parameter to be analyzed include compressor inlet pressure, compressor inlet temperature and compressor outlet pressure, and the parameter to be analyzed is compressor outlet temperature; Calculating the difference between the estimated value of the parameter to be analyzed and the actual operating data of the parameter to be analyzed to obtain the difference value of the parameter to be analyzed; Using actual operation data corresponding to the plurality of second sensor parameters as input values ​​of a compensation model, using differences of the parameters to be analyzed as sample labels, and training the compensation model; Dividing the operation process of the gas turbine into a plurality of operation stages; Determine a target second sensor parameter corresponding to each of the operation stages; For each of the operating stages, the actual operating data of the target second sensor parameter is used as the input value of the compensation model, and the difference of the parameter to be analyzed is used as a sample label to train the compensation model. The operating process of the gas turbine includes the following stages: a starting stage, a full-speed no-load stage, and a load adjustment stage. In the starting stage, the speed and the ratio of IGV to fuel flow are selected as the inputs of the compensation model. In the full-speed no-load stage and the load adjustment stage, the ratio of IGV to fuel flow and the compressor pressure ratio are selected as the inputs of the compensation model. A target compensation model corresponding to each of the operation stages is determined.

2. The method according to claim 1, wherein: The plurality of the second sensor parameters include at least one of the following: The rotation speed, the ratio of the IGV to the fuel flow rate, and the compressor pressure ratio.

3. The method according to claim 1, wherein: The mechanism model is: Wherein, p1 is the compressor inlet pressure, unit: Pa; P2 is the compressor outlet pressure, unit: Pa; T1 is the compressor inlet temperature, unit: K; T2 is the compressor outlet temperature, unit: K; is the compressor efficiency; k is the specific heat ratio of air.

4. The method according to claim 1, wherein: The compensation model includes a multi-layer neural network, and the multi-layer neural network includes: In the input layer, the number of neurons is consistent with the number of input variables; The hidden layer is a standard neuron layer or a convolutional neural network layer, the number of the hidden layers is not less than one, and the number of neurons in each hidden layer is not less than the number of the input variables; In the output layer, the number of neurons is consistent with the number of output variables.

5. The method according to claim 4, further comprising: The samples used for the compensation model training are standardized to eliminate the influence of the numerical magnitude difference of the samples on the multi-layer neural network training process, wherein the standardization includes: Obtaining the mean and standard deviation of the sample; The sample is divided by the standard deviation after subtracting the mean.

6. The method according to claim 4, wherein: Training the multi-layer neural network further includes: The stochastic gradient descent algorithm is used to gradually converge the cost function to the minimum value; In the iterative process, the back propagation algorithm is used to calculate the gradient value of the cost function based on randomly selected small batch sample data.

7. A compensation method using the compensation model according to any one of claims 1 to 4, comprising: acquiring data of a second sensor parameter of the gas turbine to be measured; Inputting the data of the second sensor parameter of the gas turbine to be measured into the trained compensation model; Outputting the compensation value of the gas turbine to be measured through the trained compensation model; The resolution margin of the second sensor parameter is obtained by comparing the compensation value with the estimated value of the parameter to be resolved.