Quantitative detection method for blockage severity of gas inlet filter screen of gas turbine
Through neural network model, the theoretical healthy pressure difference of gas turbine intake system is predicted, the severity of gas turbine intake filter blockage is quantified, and the misleading diagnosis problem caused by changes in gas working fluid in the existing technology is solved, and the complex variable working conditions are achieved Accurate blockage diagnosis.
Patent Information
- Application Number
- CN202411081056.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-12-03
AI Technical Summary
When detecting the clogged air intake filter of the gas turbine, the prior art relies solely on the pressure difference signal of the intake system, and it is easy to cause misleading diagnosis results due to changes in the mass flow rate and specific capacity of the gas working fluid, and it is impossible to accurately quantify the severity of the blockage.
By initializing the neural network, inputting environmental conditions and control conditions, establishing a pressure difference prediction model for the intake system, calculating the theoretical healthy pressure difference, and determining the health characteristic parameters based on the difference between the measured pressure difference and the theoretical pressure difference, quantifying the intake filter clogging faults. degree.
Accurately predict the theoretical health pressure difference of the intake system under complex variable operating conditions, calculate health characteristic parameters, and realize accurate quantitative diagnosis of the severity of the intake filter of the gas turbine intake filter, improving the accuracy and anti-interference of the diagnosis.
Smart Images

Figure CN119066567A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of heavy-duty gas turbine state monitoring and fault diagnosis, and in particular to a method for quantitatively detecting the severity of blockage of a gas turbine air intake filter. Background Art
[0002] At present, the health monitoring of the air intake system in a steam-gas combined cycle power plant is mainly based on the pressure difference signal of the air intake system. For example, when the pressure difference signal of the air intake system is greater than a certain monitoring threshold, it is determined that the air intake filter device needs to be replaced. However, from a mathematical point of view, the pressure difference of the air intake system is not only related to the blockage of the inlet filter, but also to the mass flow rate and specific volume of the gas working fluid. When the mass flow rate, specific volume and other parameters of the gas working fluid flowing through the pipeline change due to changes in the operating conditions of the unit, the pressure difference of the air intake system will also change accordingly. Therefore, judging the blockage of the inlet filter only from the pressure difference is too one-sided and it is easy to produce misleading diagnostic results. Summary of the invention
[0003] The purpose of the present invention is to better detect the severity of gas turbine intake filter blockage and provide a gas turbine intake filter blockage severity quantitative detection method, by setting the input of the detection model to environmental conditions (including ambient pressure P0, ambient temperature t0, relative humidity RH) and control conditions (including compressor inlet guide vane position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e ) is determined, the detection model obtains the theoretical healthy pressure difference of the intake system of the actual gas turbine unit when it is operating under complex variable operating conditions based on the above output, and then calculates the health characteristic parameters. The input selected by the present invention can well detect the pressure difference of the intake system when it is operating under complex variable operating conditions, and then calculate the health characteristic parameters to quantify the severity of the blockage of the gas turbine intake filter.
[0004] The purpose of the present invention can be achieved by the following technical solutions:
[0005] A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter, the method comprising the following steps:
[0006] S1, initializing the neural network for the gas turbine;
[0007] S2, inputting the historical gas path measurable parameters when the gas turbine is healthy into the initialized neural network for training, wherein the historical gas path measurable parameters include environmental conditions and control conditions, wherein the environmental conditions include environmental pressure P0, ambient temperature t0, and relative humidity RH, and the control conditions include compressor inlet guide vane position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e, obtain the intake system pressure difference prediction model;
[0008] S3, inputting the actual gas path measurable parameters of the gas turbine to be tested into the intake system pressure difference prediction model to obtain the theoretical healthy pressure difference of the intake system;
[0009] S4. Calculate the health characteristic parameters of the gas turbine to be tested based on the theoretical healthy pressure difference of the intake system and the measured value of the intake system pressure difference of the gas turbine to be tested, and determine the degree of blockage failure of the inlet filter of the gas turbine to be tested based on the health characteristic parameters.
[0010] Furthermore, the specific steps of inputting the historical gas path measurable parameters when the gas turbine is healthy into the initialized neural network for training are:
[0011] The historical gas path measurable parameters are input into the neural network, the loss function is calculated based on the output of the neural network and the historical intake system pressure difference, the parameters of the neural network are optimized based on the loss function, and the training is iterated until convergence.
[0012] Furthermore, the health characteristic parameters are specifically:
[0013]
[0014] Among them, ΔP IN,act is the measured value of the intake system pressure difference, ΔP IN,pre is the theoretical health pressure difference, HP IN Represents health characteristic parameters.
[0015] Furthermore, the health characteristic parameter is a dimensionless value.
[0016] Furthermore, the health characteristic parameter is greater than or equal to 1.
[0017] Furthermore, the degree of blockage of the inlet filter of the gas turbine to be tested is determined based on the health characteristic parameters as follows:
[0018] When the health characteristic parameter is equal to 1, it means that the inlet filter of the gas turbine under test has no fault;
[0019] When the health characteristic parameter is greater than 1, it indicates that the inlet filter of the gas turbine to be tested has a blockage fault.
[0020] Furthermore, the greater the absolute value of the health characteristic parameter is, the more serious the blockage fault of the inlet filter of the gas turbine to be tested is.
[0021] Furthermore, the gas turbine includes a compressor, adjustable guide vanes, a combustion chamber and a burner.
[0022] Furthermore, the measured value of the intake system pressure difference of the gas turbine to be tested is the measured value of the intake system pressure difference under complex variable operating conditions.
[0023] Furthermore, the complex variable operating condition includes the IGV minimum opening load after grid connection to the IGV fully open basic load.
[0024] Compared with the prior art, the present invention has the following beneficial effects:
[0025] The present invention sets specific training parameters of the neural network for detecting the severity of gas turbine inlet filter blockage, namely, environmental conditions (including environmental pressure P0, environmental temperature t0, relative humidity RH) and control conditions (including compressor inlet guide vane position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e ), so that the neural network can accurately predict the theoretical healthy pressure difference of the intake system under complex and variable operating conditions, and then calculate the health characteristic parameters to quantify the severity of the gas turbine intake filter blockage. The health characteristics can accurately reflect the severity of the fault. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 is a flow chart of the present invention;
[0027] Figure 2 This is a flow chart of model training of the present invention;
[0028] Figure 3 This is a graph showing changes in atmospheric temperature;
[0029] Figure 4 The opening change of the IGV of the present invention;
[0030] Figure 5 The fuel mass flow rate variation of the present invention;
[0031] Figure 6 The diagnosis result obtained by the conventional method based on the measured pressure difference signal of the intake system of the present invention;
[0032] Figure 7 It is the health characteristic parameter diagnosed when the inlet filter of the present invention is in a relatively lightly blocked condition;
[0033] Figure 8 The health characteristic parameter is obtained by diagnosing when the inlet filter of the present invention is in a moderately blocked condition;
[0034] Fig. 9 It is a health characteristic parameter diagnosed when the inlet filter of the present invention is seriously blocked. DETAILED DESCRIPTION
[0035] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.
[0036] In view of the problem that the existing technology only judges the blockage status of the inlet filter of the air intake system from the pressure difference, which is too one-sided, the present invention proposes a method for quantitatively detecting the severity of the blockage of the gas turbine air intake filter, using the healthy characteristic parameters of the air intake system obtained by diagnosis as a quantitative indicator to characterize the blockage status of the inlet filter of the air intake system, and diagnoses the severity of the blockage of the air intake filter of the air intake system from the mathematical essence, providing a scientific basis for formulating appropriate and reasonable maintenance strategies. The flow chart of the method proposed by the present invention is as follows Figure 1 shown.
[0037] The method of the present invention comprises the following steps:
[0038] S1, initializing the neural network for the gas turbine;
[0039] S2, inputting the historical gas path measurable parameters when the gas turbine is healthy into the initialized neural network for training, wherein the historical gas path measurable parameters include environmental conditions and control conditions, wherein the environmental conditions include environmental pressure P0, ambient temperature t0, and relative humidity RH, and the control conditions include compressor inlet guide vane position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e , obtain the intake system pressure difference prediction model;
[0040] S3, inputting the actual gas path measurable parameters of the gas turbine to be tested into the intake system pressure difference prediction model to obtain the theoretical healthy pressure difference of the intake system;
[0041] S4. Calculate the health characteristic parameters of the gas turbine to be tested based on the theoretical healthy pressure difference of the intake system and the measured value of the intake system pressure difference of the gas turbine to be tested, and determine the degree of blockage failure of the inlet filter of the gas turbine to be tested based on the health characteristic parameters.
[0042] In the present invention, the historical gas path measurable parameters of the gas turbine intake system to be diagnosed when it is operating normally and healthily are first collected, and an intake system pressure difference prediction model is established through a deep neural network. The intake system pressure difference prediction model is used to calculate in real time the theoretical healthy pressure difference of the intake system of the actual gas turbine unit when it is operating under complex variable operating conditions; the intake system health characteristic parameters are designed as quantitative indicators to characterize the blockage condition of the intake system inlet filter. The health characteristic parameter is equal to the difference between the measured pressure difference and the theoretical pressure difference of the intake system divided by the theoretical pressure difference, plus a bias value of 1. The dimensionless health characteristic parameter obtained by the diagnosis reflects the physical nature of the change in the inlet filter blockage condition, and can maintain good robustness and accuracy at different variable operating points. When the value of the health characteristic parameter is 1, it means that the intake system is in a baseline health state, and when the value is greater than 1, it means that there is an inlet filter blockage condition, and the larger the value, the more serious it is. Through the method of the present invention, the inlet filter blockage condition of the gas turbine intake system can be accurately and quantitatively diagnosed under complex variable operating conditions, providing a scientific basis and technical support for formulating appropriate and reasonable maintenance strategies. The model training process of the present invention is as follows Figure 2 shown.
[0043] The historical gas path measurable parameters of the gas turbine intake system under normal and healthy operation are collected to establish an intake system pressure difference prediction model through a deep neural network. For an actual healthy heavy-duty gas turbine, its operating performance status (including intake system pressure difference) can be uniquely determined by environmental conditions (including ambient pressure P0, ambient temperature t0, relative humidity RH) and control conditions (including compressor inlet guide vane position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e ), so the measurable parameters of the gas path input by the deep neural network include environmental conditions (including ambient pressure P0, ambient temperature t0, relative humidity RH) and control conditions (including IGV position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e ), the measurable gas path parameter output by the deep neural network is the intake system pressure difference ΔP IN (i.e. the difference between the ambient pressure P0 and the compressor inlet total pressure P1). The intake system pressure difference prediction model is used to calculate the theoretical pressure difference ΔP of the intake system of the actual gas turbine unit under complex variable operating conditions in real time. IN,pre .
[0044] The health characteristic parameter HP IN Equal to the measured pressure difference of the intake system ΔP IN,act The theoretical healthy pressure difference ΔP IN,pre The difference is divided by the theoretical healthy pressure difference ΔP IN,pre , plus a bias value of 1, that is:
[0045]
[0046] The dimensionless health characteristic parameter obtained by the diagnosis reflects the physical nature of the change in the inlet filter blockage condition, and can maintain good robustness and accuracy at different operating points. IN When the value is 1, it indicates that the intake system is in a baseline healthy state. When the value is greater than 1, it indicates that the inlet filter is clogged, and the larger the value, the more serious the condition.
[0047] In order to test the effectiveness of the diagnostic method proposed in the present invention, the AE94.3AF heavy-duty gas turbine is selected as the diagnostic object. This AE94.3A heavy-duty gas turbine has been widely used in domestic power plants. Its compressor is a 15-stage large-flow, axial-flow compressor equipped with 2-stage inlet adjustable guide vanes. The combustion chamber is an annular combustion chamber equipped with 24 low-NOx burners. The detailed gas path measurable parameters of this heavy-duty gas turbine are shown in Table 1.
[0048] Table 1 Environmental and control parameter conditions and other measurable parameters of the gas path
[0049]
[0050] First, the historical measurable gas path parameters of the gas turbine intake system to be diagnosed when it was healthy are collected, and a pressure differential prediction model for the intake system is established through a deep neural network. The historical measurable gas path parameters of the gas turbine intake system to be diagnosed when it was operating normally and healthily are collected, and a pressure differential prediction model for the intake system is established through a deep neural network. For an actual healthy heavy-duty gas turbine, its operating performance status (including the intake system pressure differential) can be uniquely determined by environmental conditions (including ambient pressure P0, ambient temperature t0, relative humidity RH) and control conditions (including compressor inlet guide vane position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e ), so the measurable parameters of the gas path input by the deep neural network include environmental conditions (including ambient pressure P0, ambient temperature t0, relative humidity RH) and control conditions (including IGV position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e ), the measurable gas path parameter output by the deep neural network is the intake system pressure difference ΔP IN (i.e. the difference between the ambient pressure P0 and the compressor inlet total pressure P1). The intake system pressure difference prediction model is used to calculate in real time the theoretical healthy pressure difference ΔP of the intake system of the actual gas turbine unit under complex variable operating conditions. IN,pre .
[0051] The intake system health characteristic parameter is then designed as a quantitative indicator to characterize the blockage condition of the intake system inlet filter. IN Equal to the measured pressure difference of the intake system ΔP IN,act The theoretical healthy pressure difference ΔP IN,pre The difference is divided by the theoretical pressure difference ΔP IN,pre , plus a bias value of 1, that is:
[0052]
[0053] The dimensionless health characteristic parameters obtained by this diagnosis reflect the physical essential changes in the inlet filter blockage condition and can maintain good robustness and accuracy at different operating points.
[0054] In order to test the effectiveness of the proposed method in the present invention in accurately quantifying the clogging condition of the inlet filter of the gas turbine intake system under complex variable operating conditions (from the minimum opening load of the IGV after grid connection to the basic load of the IGV fully opened), the heavy-duty gas turbine unit was selected under complex variable operating conditions (such as Figures 3 to 5 The diagnostic test is performed on the samples of measurable parameters of the air path when the inlet filter of the air intake system is blocked with different severity (mild, medium, severe) as shown in Table 2. The description of the blockage of the inlet filter of the air intake system with different severity is shown in Table 2.
[0055] Table 2 Description of the blockage of the inlet filter of the intake system with different degrees of severity
[0056]
[0057] The results of the traditional diagnosis method based on the measured pressure difference signal of the intake system are as follows Figure 6 shown.
[0058] Depend on Figure 6 As shown, since the pressure difference of the intake system is not only related to the blockage of the inlet filter, but also to the mass flow rate and specific volume of the gas medium, when the mass flow rate and specific volume of the gas medium flowing through the pipeline change due to the change of the unit operating conditions, the pressure difference of the intake system will also change accordingly. When the inlet filter of the intake system is blocked to different degrees, the pressure difference of the intake system fluctuates and overlaps greatly at different variable operating points. Therefore, the existing technology is too one-sided to judge the condition of the pipeline only from the pressure difference, which is easy to produce misleading diagnostic results.
[0059] The results of the diagnosis obtained by the method proposed by the present invention are as follows Figures 7 to 9 shown.
[0060] Depend on Figures 7 to 9As shown, the quantitative numerical results diagnosed by the method proposed in the present invention are basically consistent with the blockage conditions of different severity (mild, medium, severe) of the inlet filter of the intake system of the heavy-duty gas turbine unit under complex variable operating conditions, and the diagnosed health characteristic parameters of the intake system maintain good robustness and accuracy at different variable operating points, and are not affected by changes in the operating conditions of the unit, which shows that the method proposed in the present invention has higher diagnostic accuracy and anti-interference ability.
[0061] For an actual healthy heavy-duty gas turbine, its operating performance state (including the intake system pressure difference) can be determined solely by the environmental conditions (including ambient pressure P0, ambient temperature t0, relative humidity RH) and control conditions (including compressor inlet guide vane position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e ), so the measurable parameters of the gas path input by the deep neural network include environmental conditions (including ambient pressure P0, ambient temperature t0, relative humidity RH) and control conditions (including IGV position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e ), the measurable gas path parameter output by the deep neural network is the intake system pressure difference ΔP IN (i.e. the difference between the ambient pressure P0 and the compressor inlet total pressure P1). The intake system pressure difference prediction model is used to calculate in real time the theoretical healthy pressure difference ΔP of the intake system of the actual gas turbine unit under complex variable operating conditions. IN,pre Next, we further analyzed the sensitivity of each input parameter of the selected deep neural network to the diagnostic accuracy, as shown in Table 3.
[0062] Table 3 Sensitivity analysis of the selected deep neural network input parameters on diagnostic accuracy
[0063]
[0064] It can be seen from Table 3 that when the heavy-duty gas turbine is running under variable conditions, among the selected input parameters of the deep neural network, the compressor inlet guide vane position feedback IGV is the most sensitive to the diagnostic accuracy, followed by the ambient temperature t0 and the ambient pressure P0, and then the gas turbine speed n, while the relative humidity RH and the gas turbine power N e and fuel temperature t f The sensitivity to diagnostic accuracy is relatively small (the maximum error percentage does not exceed 0.6%), so the present invention can further select the deep neural network input parameters for modeling the intake system pressure difference prediction model as four parameters including environmental conditions (including ambient pressure P0, ambient temperature t0) and control conditions (including IGV position feedback IGV, engine speed n).
[0065] It can be seen from the above specific implementation cases that the method proposed in the present invention can accurately quantify the blockage condition of the inlet filter of the gas turbine intake system under complex and variable operating conditions, providing a scientific basis and technical support for formulating appropriate and reasonable maintenance strategies.
[0066] Since the pressure difference of the air intake system is not only related to the blockage of the inlet filter, but also to the mass flow rate and specific volume of the gas medium, when the mass flow rate and specific volume of the gas medium flowing through the pipeline change due to the change of the unit's operating conditions, the pressure difference of the air intake system will also change accordingly. When the inlet filter of the air intake system is blocked to varying degrees, the pressure difference of the air intake system will fluctuate and overlap greatly at different operating points under different conditions. Therefore, the existing technology is too one-sided to judge the condition of the pipeline only from the pressure difference, which is easy to produce misleading diagnostic results.
[0067] In order to effectively extract the healthy characteristic parameters of the intake system that are not affected by changes in the unit's operating conditions, the present invention first collects the historical gas path measurable parameters when the intake system of the gas turbine to be diagnosed is healthy, and establishes an intake system pressure difference prediction model through a deep neural network. The intake system pressure difference prediction model is then used to calculate in real time the theoretical pressure difference ΔP of the intake system of the actual gas turbine unit when it is operating under complex variable conditions. IN,pre Finally, the intake system health characteristic parameter is designed as a quantitative indicator to characterize the blockage condition of the intake system inlet filter. IN Equal to the measured pressure difference of the intake system ΔP IN,act The theoretical pressure difference ΔP IN,pre The difference is divided by the theoretical pressure difference ΔP IN,pre , plus the bias value 1, the dimensionless health characteristic parameters obtained by this diagnosis reflect the physical essential changes in the inlet filter blockage condition, and can maintain good robustness and accuracy at different operating points.
[0068] The method of the present invention can accurately quantify the clogging condition of the inlet filter of the gas turbine intake system under complex and variable operating conditions, providing a scientific basis and technical support for formulating appropriate and reasonable maintenance strategies.
[0069] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.
Claims
1. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter, characterized in that: The method comprises the following steps: S1, initializing the neural network for the gas turbine; S2, inputting the historical gas path measurable parameters when the gas turbine is healthy into the initialized neural network for training, wherein the historical gas path measurable parameters include environmental conditions and control conditions, wherein the environmental conditions include environmental pressure P0, ambient temperature t0, and relative humidity RH, and the control conditions include compressor inlet guide vane position feedback IGV, fuel temperature t f , gas turbine speed n, gas turbine power N e , obtain the intake system pressure difference prediction model; S3, inputting the actual gas path measurable parameters of the gas turbine to be tested into the intake system pressure difference prediction model to obtain the theoretical healthy pressure difference of the intake system; S4. Calculate the health characteristic parameters of the gas turbine to be tested based on the theoretical healthy pressure difference of the intake system and the measured value of the intake system pressure difference of the gas turbine to be tested, and determine the degree of blockage failure of the inlet filter of the gas turbine to be tested based on the health characteristic parameters.
2. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter according to claim 1, characterized in that: The specific steps of inputting the historical gas path measurable parameters when the gas turbine is healthy into the initialized neural network for training are: The historical gas path measurable parameters are input into the neural network, the loss function is calculated based on the output of the neural network and the historical intake system pressure difference, the parameters of the neural network are optimized based on the loss function, and the training is iterated until convergence.
3. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter according to claim 2, characterized in that: The health characteristic parameters are specifically: Where ΔP IN,act is the measured value of the intake system pressure difference, ΔP IN,pre is the theoretical health pressure difference, HP IN Represents health characteristic parameters.
4. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter according to claim 3, characterized in that: The health characteristic parameter is a dimensionless value.
5. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter according to claim 4, characterized in that: The health characteristic parameter is greater than or equal to 1.
6. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter according to claim 5, characterized in that: The specific degree of blockage of the inlet filter of the gas turbine to be tested is determined based on the health characteristic parameters: When the health characteristic parameter is equal to 1, it means that the inlet filter of the gas turbine under test has no fault; When the health characteristic parameter is greater than 1, it indicates that the inlet filter of the gas turbine to be tested has a blockage fault.
7. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter according to claim 6, characterized in that: The larger the absolute value of the health characteristic parameter is, the more serious the blockage failure of the inlet filter of the gas turbine to be tested is.
8. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter according to claim 1, characterized in that: The gas turbine comprises a compressor, adjustable guide vanes, a combustion chamber and a burner.
9. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter according to claim 1, characterized in that: The measured value of the intake system pressure difference of the gas turbine to be tested is the measured value of the intake system pressure difference when the gas turbine is running under complex variable operating conditions.
10. A method for quantitatively detecting the severity of blockage of a gas turbine air intake filter according to claim 1, characterized in that: The complex variable operating conditions include the minimum opening load of the IGV after grid connection to the basic load of the IGV fully open.
