Method and system for predicting combustion temperature of gas turbine

Abstract

The invention discloses a gas turbine combustion temperature prediction method and system, and relates to the field of gas turbine combustion temperature monitoring and prediction, and the method comprises the steps: obtaining combustion temperature data, constructing a data set, and constructing an envelope line; performing first judgment based on the envelope line, and if not, performing reconstruction until a judgment result is met; eMD decomposition is carried out on the original signal by using the envelope line after the first judgment is terminated, and a data set is divided according to a decomposition result to obtain a preliminary division result; performing recursive division on the preliminary division result, obtaining a final division result after a preset end condition is met, and further constructing a decision tree for combustion temperature prediction; aiming at the problems of high cost, slow sampling, limited range and the like of traditional temperature measurement, big data and machine learning are combined, empirical mode decomposition is applied, a primary function does not need to be preset, a decision tree regression algorithm is adopted for modeling prediction, and the method has the advantages of being good in universality, low in cost and low in time complexity.

Description

Technical Field

[0001] The present invention relates to the field of gas turbine combustion temperature monitoring and prediction, and in particular to a gas turbine combustion temperature prediction method and system. Background Art

[0002] Combustion temperature is a concept that involves multiple factors, including fuel type, combustion conditions, and heat transfer. Combustion temperature significantly impacts engine performance, fuel economy, and emissions. Excessively high combustion temperatures primarily impact gas turbines by damaging combustion chamber materials, reducing power generation efficiency, and increasing energy consumption. The combustion chamber in a gas turbine is a key component that converts the chemical energy of the fuel into thermal energy. Its extremely high operating temperatures place stringent demands on material properties. Increased combustion temperatures pose even greater challenges to material properties, particularly requiring materials to possess high-temperature oxidation resistance and gas corrosion resistance, sufficient transient performance and sustained strength, good thermal fatigue resistance, and excellent processability. Excessively high compressor outlet temperatures transfer more heat to the air, leading to excessively high temperatures within the combustion chamber and damaging the combustion chamber materials. Furthermore, excessive temperatures can affect gas turbine performance, reducing power generation efficiency and increasing energy consumption. Gas turbine efficiency is closely related to its inlet air temperature; higher temperatures result in lower efficiency. This is because high temperatures can cause deformation of the compressor and turbine blades within the gas turbine, reducing efficiency. In addition, high temperatures can cause instability in the combustion process within the gas turbine, affecting the complete combustion of the fuel and thus reducing power generation efficiency. Therefore, controlling the combustion temperature of the gas turbine within an appropriate range is an important measure to ensure the safe and reliable operation of the gas turbine.

[0003] The combustion temperature monitoring methods commonly used in engineering generally rely on sensor measurements, but they have disadvantages such as high cost, slow sampling rate, limited temperature measurement range, significant influence from environmental factors, and limited measurement distance. Existing theoretical prediction methods for combustion temperature are mostly based on principle modeling, which has complex model construction, difficult implementation, and low universality. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are: the traditional gas turbine combustion temperature measurement method has high cost, slow sampling rate, limited temperature measurement range, and the existing theoretical prediction method model is complex to build, difficult to implement, and has low universality.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for predicting combustion temperature of a gas turbine, comprising:

[0008] Obtain combustion temperature data to construct a data set, and construct an envelope based on the data set;

[0009] Perform a first judgment based on the envelope, terminate the first judgment if the judgment result is satisfied, and reconstruct the envelope if the judgment result is not satisfied until the judgment result is satisfied;

[0010] Based on the envelope after the first judgment is terminated, the EMD decomposition of the original signal is performed, and the data set is subsetted according to the decomposition result to obtain a preliminary partitioning result;

[0011] Based on the preliminary division results, recursive division is performed until the preset recursive end condition is reached to obtain the final division result; a decision tree is constructed based on the final division result, and the combustion temperature of the gas turbine is predicted using the decision tree.

[0012] As a preferred method for predicting the combustion temperature of a gas turbine, the following is a preferred method:

[0013] The constructing of the envelope based on the data set includes:

[0014] The envelope is constructed using the interpolation method. An upper envelope in the signal connects the maximum points of the interpolation points, and a lower envelope in the signal connects the minimum points of the interpolation points.

[0015] As a preferred method for predicting the combustion temperature of a gas turbine, the following is a preferred method:

[0016] The step of performing a first judgment based on the envelope, terminating the first judgment if the judgment result is satisfied, and reconstructing the envelope if the judgment result is not satisfied until the judgment result is satisfied includes:

[0017] The first-order intrinsic mode function is iteratively separated from the original signal of the data set. In each iteration, the signal is made to meet the conditions of the intrinsic mode function by calculating the average function and the difference. When the conditions of the intrinsic mode function are met, the iteration is stopped; otherwise, the current difference signal is used as the new original signal and the iteration is continued until the conditions of the intrinsic mode function are met.

[0018] As a preferred method for predicting the combustion temperature of a gas turbine, the following is a preferred method:

[0019] The data set is divided into subsets according to the decomposition result to obtain preliminary division results, including:

[0020] Calculate the mean square error of the EMD decomposition result of the original signal. If the mean square error is greater than the set value, it is classified as abnormal data, otherwise it is normal data.

[0021] As a preferred method for predicting the combustion temperature of a gas turbine, the following is a preferred method:

[0022] The recursive partitioning is performed based on the preliminary partitioning result until a preset recursive termination condition is reached to obtain a final partitioning result; and constructing a decision tree based on the final partitioning result includes:

[0023] Recursive partitioning: recursive partitioning is performed based on the preliminary partitioning results. Each partitioning determines the output of the partitioning unit and adds a new node. A heuristic method is used to partition the feature space. Each time, all values ​​of all features in the current set are examined, and the optimal split point is selected according to the square error minimization criterion.

[0024] As a preferred method for predicting the combustion temperature of a gas turbine, the following is a preferred method:

[0025] The method further comprises: performing recursive partitioning based on the preliminary partitioning result until a preset recursive termination condition is reached to obtain a final partitioning result; and constructing a decision tree based on the final partitioning result further comprises:

[0026] Select the split point: examine the features and their values ​​one by one, traverse the feature variables, scan the split points for the fixed features, and select the feature and split point combination that minimizes the square error;

[0027] Divide the region: Divide the region by combining the selected features and the split points, and determine the output value of the corresponding region;

[0028] Repeated division: Repeat the steps of selecting the split point and dividing the area for the two sub-areas until the preset stop condition is met;

[0029] The input space is divided into multiple regions, and finally a decision tree is generated, where each region corresponds to a leaf node output value.

[0030] As a preferred method for predicting the combustion temperature of a gas turbine, the following is a preferred method:

[0031] The method of using a decision tree to predict the combustion temperature of a gas turbine includes:

[0032] The input sample is traversed along the path of the decision tree until it reaches a leaf node, and the output value of the leaf node is used as the prediction result.

[0033] In a second aspect, an embodiment of the present invention provides a gas turbine combustion temperature prediction system, comprising:

[0034] An envelope curve construction module is used to obtain combustion temperature data to construct a data set, and to construct an envelope curve based on the data set;

[0035] A first judgment module is configured to perform a first judgment based on the envelope, terminate the first judgment if the judgment result is satisfied, and reconstruct the envelope if the judgment result is not satisfied until the judgment result is satisfied;

[0036] A preliminary partitioning module is used to perform EMD decomposition of the original signal based on the envelope after the first judgment is terminated, and to subset the data set according to the decomposition result to obtain a preliminary partitioning result;

[0037] The decision tree prediction module is used to perform recursive partitioning based on the preliminary partitioning results until the preset recursive end condition is reached to obtain the final partitioning result; a decision tree is constructed based on the final partitioning result, and the decision tree is used to predict the combustion temperature of the gas turbine.

[0038] In a third aspect, an embodiment of the present invention provides a computing device, including:

[0039] memory and processor;

[0040] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement the gas turbine combustion temperature prediction method as described in any embodiment of the present invention.

[0041] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which implement the gas turbine combustion temperature prediction method when executed by a processor.

[0042] The beneficial effects of the present invention are as follows: In response to the problems of high cost, slow sampling rate and limited temperature measurement range in traditional temperature measurement methods, the present invention proposes a combustion temperature prediction method based on big data methods and machine learning, which saves engineering application costs and improves the universality of the method; the empirical mode decomposition method is used to decompose the signal according to the time scale characteristics of the historical temperature data itself, without the need to pre-set any basis functions; the decision tree regression algorithm is used to classify, model and predict the data after empirical mode decomposition; compared with the existing technology, the combustion temperature prediction method proposed by the present invention has good universality, low cost and low time complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 It is an overall flow chart of the gas turbine combustion temperature prediction method according to the present invention;

[0045] Figure 21 is a schematic diagram of prediction results of the combustion temperature prediction method in a simulation example of the gas turbine combustion temperature prediction method according to the present invention;

[0046] Figure 3 It is a combustion temperature prediction error curve diagram in a simulation example of the gas turbine combustion temperature prediction method described in the present invention. DETAILED DESCRIPTION

[0047] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0048] Example 1, reference Figure 1 , which is a first embodiment of the present invention, provides a method for predicting combustion temperature of a gas turbine, comprising:

[0049] S1: Obtain combustion temperature data to construct a data set, and construct an envelope based on the data set;

[0050] S2: Perform a first judgment based on the envelope line. If the judgment result is satisfied, terminate the first judgment. If the judgment result is not satisfied, reconstruct the envelope line until the judgment result is satisfied.

[0051] S3: Based on the envelope after the first judgment is terminated, perform EMD decomposition of the original signal, and subset the data set according to the decomposition result to obtain a preliminary partitioning result;

[0052] S4: Based on the preliminary division results, recursive division is performed until a preset recursive end condition is reached to obtain a final division result; a decision tree is constructed based on the final division result, and the combustion temperature of the gas turbine is predicted using the decision tree.

[0053] It should be noted that the gas turbine combustion temperature prediction method based on EMD and decision tree regression, constructed through steps S1-S4 above, can effectively and accurately predict gas turbine combustion temperature. Using EMD to decompose the raw combustion temperature data, the complex original signal is decomposed into multiple intrinsic mode functions (IMFs) with different time scales. This decomposition method can deeply explore the local characteristics and different frequency components in the signal, making the data characteristics clearer and easier to analyze. Furthermore, during the decision tree construction process, by continuously partitioning and recursively processing the dataset, the optimal partitioning attributes are selected based on the inherent laws of the data, allowing the decision tree to fully learn the characteristics and patterns of the data. This method combines the advantages of EMD in signal processing with the capabilities of decision trees in data classification and regression. Compared with traditional single prediction methods, it can better adapt to the complexity and nonlinear characteristics of gas turbine combustion temperature data, reduce the impact of noise and outliers on prediction results, and improve prediction accuracy and stability.

[0054] Example 2, reference Figure 1 , which is an embodiment of the present invention, provides a method for predicting combustion temperature of a gas turbine based on the previous embodiment, comprising:

[0055] In this embodiment, the step S1 of acquiring the combustion temperature data and constructing the data set includes:

[0056] Obtain historical combustion temperature data x(t), perform preprocessing operations such as cleaning and removing bad values ​​on the data to obtain a data set;

[0057] Find the maximum and minimum points of the data in the data set, and divide the original signal into a series of small segments through the maximum and minimum points;

[0058] Interpolation is performed on each small segment so that evenly distributed data points can fit this small segment.

[0059] In this embodiment, constructing the envelope based on the data set in step S1 includes:

[0060] Use the Cubic Spline interpolation method or other interpolation methods to construct an envelope, which connects the maximum and minimum points of the interpolation points, that is, an upper envelope and a lower envelope in the signal;

[0061] Specifically, all the maximum points are interpolated with the cubic spline function curve to fit the upper envelope x of the original signal x(t) max (t). Similarly, the lower envelope x of the data set x(t) (i.e. the original signal) can also be obtained min (t). The upper and lower envelopes contain all the data information of the signal.

[0062] It should be noted that the selected interpolation method is used to process the maximum points. This interpolation method creates a continuous curve connecting all the maximum points, which is the upper envelope of the signal. The upper envelope always lies above the original data x(t) throughout the entire data time range, reflecting the maximum trend of the original signal in each local area.

[0063] Similarly, the selected interpolation method is applied to the minimum points. This interpolation method generates a continuous curve connecting all the minimum points, which is the lower envelope of the signal. The lower envelope always lies below the original data x(t) throughout the entire data time range, reflecting the trend of the original signal's minimum values ​​in each local region.

[0064] In this embodiment, the first judgment is performed based on the envelope in the above step S2. If the judgment result is satisfied, the first judgment is terminated. If the judgment result is not satisfied, the envelope is reconstructed until the judgment result is satisfied.

[0065] The specific steps of the first judgment are:

[0066] Calculate the average function: calculate the average function m(t) = (upper envelope + lower envelope) / 2;

[0067] Specifically, the mean values ​​of the upper and lower envelopes are connected in sequence to obtain a mean line m1(t):

[0068] m1(t)=[x max (t)+x min (t)] / 2

[0069] Calculate the difference: Calculate the difference d(t) between x(t) and m(t), d(t) = x(t) - m(t);

[0070] Specifically, we can get h1(t) by subtracting m1(t) from x(t), which is expressed as:

[0071] h1(t)=x(t)-m1(t)

[0072] It should be noted that this process, called "sieving," has two objectives: eliminating riding waves and making the waveform more symmetrical. Eliminating riding waves primarily ensures that meaningful instantaneous frequencies are obtained when performing the Hilbert transform, while the second objective is to prevent large amplitude differences between adjacent waves. To achieve this result, the sieving process must be repeated multiple times. In subsequent sieving steps, h1(t) is treated as data again, and the above steps are repeated.

[0073] h 11 (t)=h1(t)-m11 (t)

[0074] When this process is repeated k times, h 1k (t) Until the two conditions of the intrinsic mode function are met, then:

[0075] h 1k (t) = h 1(k-1) (t)-m 1k (t)

[0076] At this time, record h 1k (t) is the first-order IMF, c1(t)=h 1k (t), c1(t) represents the shortest periodic perturbation in the observed signal x(t).

[0077] If d(t) is an IMF (Intrinsic Mode Function), the first judgment is terminated; otherwise, d(t) is used as a new original signal, and the envelope is reconstructed based on the new original signal.

[0078] Specifically, calculate the margin r1(t): r1(t)=x(t)-c1(t);

[0079] Since r1(t) still contains the longer period oscillations in the observed signal, r1(t) is treated as a new signal and the envelope is reconstructed until the margin r n (t) When it becomes a monotonic function or no more monotonic functions can be extracted, the first judgment is terminated.

[0080] In another possible implementation, the first judgment may also be a spectral feature judgment: the difference signal d(t) is subjected to spectral analysis, such as a fast Fourier transform (FFT), to determine whether its spectrum meets the characteristics of an IMF. An IMF typically has a single frequency component, so whether d(t) is an IMF can be determined by analyzing the main frequency component and frequency distribution range in the spectrum. If there are multiple obvious frequency peaks in the spectrum, or the frequency distribution is too dispersed, d(t) is considered not to be an IMF.

[0081] In another possible implementation, the first determination can also be a correlation determination: the correlation between the difference signal d(t) and the original signal x(t) and the previously filtered IMF components is calculated. If d(t) has a high correlation with the original signal or the previously filtered IMF components, it indicates that d(t) may still contain previously separated components and does not meet the IMF independence requirement, requiring the envelope to be reconstructed for screening.

[0082] In another possible implementation, the first judgment may also be an energy distribution judgment: analyzing the energy distribution of the difference signal d(t). IMFs typically have a concentrated energy distribution, meaning that the energy is primarily concentrated in one or several specific frequency bands. By calculating the energy proportion of d(t) in different frequency bands, it can be determined whether its energy distribution meets the characteristics of an IMF. If the energy distribution is too dispersed, d(t) is considered not to be an IMF.

[0083] In this embodiment, performing EMD decomposition of the original signal based on the envelope after the first judgment is terminated in step S3 includes:

[0084] The calculated IMFs are combined to obtain the EMD decomposition result of the original signal x(t).

[0085] Specifically, after EMD decomposition, the original signal can be expressed as:

[0086]

[0087] Where c j (t) is the jth IMF; r n (t) is the residual fraction after multiple screenings; n is the total number of IMFs.

[0088] From the decomposition process, it can be seen that EMD is completely adaptive based on the observed data, unlike the wavelet transform which requires the preselection of the wavelet basis function. Each intrinsic mode function component is decomposed in order from small to large scale, i.e. c j (t) Arrange in order from high to low frequency.

[0089] It should be noted that the Empirical Mode Decomposition (EMD) method is considered a major breakthrough in Fourier transform-based linear and stationary spectrum analysis since 2000. This method decomposes signals based on the time-scale characteristics of the data itself, without requiring any predefined basis functions. This fundamentally differs from Fourier and wavelet decomposition methods, which rely on a priori harmonic and wavelet basis functions. Precisely because of this characteristic, the EMD method can theoretically be applied to decompose any type of signal, offering significant advantages in processing nonstationary and nonlinear data. It is suitable for analyzing nonlinear and nonstationary signal sequences and achieves a high signal-to-noise ratio. The key to this method is empirical mode decomposition, which decomposes complex signals into a finite number of intrinsic mode functions (IMFs). Each IMF component contains local characteristic signals of the original signal at different time scales. EMD can stabilize nonstationary data, then perform a Hilbert transform to obtain a time-frequency spectrum, yielding physically meaningful frequencies.

[0090] An eigenmode function must satisfy the following two conditions: the number of local extreme points and zero crossing points in the entire time range of the function must be equal, or differ by at most one; at any point in time, the average of the envelope of the local maximum value (upper envelope) and the envelope of the local minimum value (lower envelope) must be zero.

[0091] In this embodiment, in step S3, the dataset is divided into subsets according to the decomposition result, and the preliminary division results obtained include:

[0092] The mean square error of the EMD decomposition result of the original signal is calculated. If the mean square error is greater than a set value, preferably 0.5 in this embodiment, it is classified as abnormal data; otherwise, it is normal data.

[0093] In this embodiment, in step S4, recursive partitioning is performed based on the preliminary partitioning result until a preset recursive termination condition is reached to obtain a final partitioning result; and constructing a decision tree based on the final partitioning result includes:

[0094] It should be noted that data is modeled and predicted based on a tree structure, and the regression task is achieved by dividing the dataset into different regions and predicting a constant value in each region. Decision tree regression models and predicts data by constructing a tree structure. Each internal node of the tree represents an attribute / feature, and each leaf node represents an output value. The decision tree construction process is a recursive process that selects the best attributes / features to partition the data so that the output values ​​of the partitioned subsets are as close to the true values ​​as possible. During the prediction stage, the decision tree traverses the input sample along the tree path until it reaches a leaf node, and then uses the output value of the leaf node as the prediction result.

[0095] Specifically, the values ​​of the internal node features are "yes" and "no", which is a binary tree structure. The partitioning process is also the process of building a tree. Each time a partition is made, the output corresponding to the partition unit is determined, and an additional node is added. When the partition is terminated according to the stopping condition, the output of each unit is finally determined, that is, the leaf node. Assume that X and Y are input and output variables respectively, and Y is a continuous variable. Given a training data set D = {(x1, y1), (x2, y2), ..., (x N ,y N )}, where x i =(x i (1) ,x i (2) ,...,x i (n) ) is the input instance, n is the number of features, i = 1, 2, ..., N, N is the number of samples.

[0096] The feature space is divided into two parts using a heuristic method. Each time the feature space is divided, all the values ​​of all the features in the current set are examined one by one, and the best one is selected as the dividing point according to the square error minimization criterion.

[0097] Specifically, for the j-th feature variable x in the training set (j) and its value s, as the segmentation variable and segmentation point, and define two regions R1(j,s)={x|x (j) ≤s} and R2(j,s)={x|x (j) >s}, find the optimal j and s;

[0098] Select the split point: Select the optimal split variable j and split point s, and solve:

[0099]

[0100] Where c1 and c2 are the fixed output values ​​within the two regions after partitioning. The two min values ​​in [] indicate that the optimal c1 and c2 are used, i.e., the c1 and c2 that minimize the squared error within their respective regions. We traverse variable j, scanning the split point s for the fixed split variable j, and select the pair (j, s) that minimizes the above equation.

[0101] Divide the region: Divide the region by the selected pair (j, s) and determine the corresponding output value:

[0102]

[0103] Repeated division: Repeat the steps of selecting the split point and dividing the area for the two sub-areas until the preset stop condition is met;

[0104] Divide the input space into M regions R1, R2, ..., R M , generate a decision tree:

[0105]

[0106] Where I is the indicator function,

[0107] In this embodiment, the prediction of the combustion temperature of the gas turbine using the decision tree in step S4 includes:

[0108] The input sample is traversed along the tree path and finally reaches a leaf node, and the output value of the leaf node is used as the prediction result.

[0109] Example 3. The above is a schematic diagram of the gas turbine combustion temperature prediction method of this embodiment. It should be noted that the technical solution of the gas turbine combustion temperature prediction system and the technical solution of the above-mentioned gas turbine combustion temperature prediction method are based on the same concept. For details not described in detail in the technical solution of the gas turbine combustion temperature prediction system in this embodiment, please refer to the description of the technical solution of the above-mentioned gas turbine combustion temperature prediction method.

[0110] This embodiment further provides a system based on a gas turbine combustion temperature prediction method, comprising:

[0111] An envelope curve construction module is used to obtain combustion temperature data to construct a data set, and to construct an envelope curve based on the data set;

[0112] A first judgment module is configured to perform a first judgment based on the envelope, terminate the first judgment if the judgment result is satisfied, and reconstruct the envelope if the judgment result is not satisfied until the judgment result is satisfied;

[0113] A preliminary partitioning module is used to perform EMD decomposition of the original signal based on the envelope after the first judgment is terminated, and to subset the data set according to the decomposition result to obtain a preliminary partitioning result;

[0114] The decision tree prediction module is used to perform recursive partitioning based on the preliminary partitioning results until the preset recursive end condition is reached to obtain the final partitioning result; a decision tree is constructed based on the final partitioning result, and the decision tree is used to predict the combustion temperature of the gas turbine.

[0115] This embodiment further provides a computing device applicable to the method for predicting combustion temperature of a gas turbine, comprising:

[0116] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the gas turbine combustion temperature prediction method proposed in the above embodiment.

[0117] This embodiment further provides a storage medium storing a computer program, which, when executed by a processor, implements the gas turbine combustion temperature prediction method proposed in the above embodiment.

[0118] The storage medium proposed in this embodiment and the gas turbine combustion temperature prediction method proposed in the above embodiment belong to the same inventive concept. Technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0119] Example 4, with reference to Figure 2-Figure 3 , which is an embodiment of the present invention, provides a method for predicting the combustion temperature of a gas turbine. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0120] Figure 2 The prediction results obtained by the combustion temperature prediction method proposed by the present invention are as follows: Figure 3 The combustion temperature prediction error diagram is obtained. Figure 2 It can be seen that the combustion temperature prediction results of the method proposed in this invention are basically consistent with the measured results. Figure 3 It can be seen that the prediction accuracy of the method proposed in the present invention is greater than 95%, which meets the actual requirements.

[0121] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting combustion temperature of a gas turbine, characterized in that: include: Obtain combustion temperature data to construct a data set, and construct an envelope based on the data set; Perform a first judgment based on the envelope, and terminate the first judgment if the judgment result is satisfied; if the judgment result is not satisfied, reconstruct the envelope until the judgment result is satisfied; Based on the envelope after the first judgment is terminated, the EMD decomposition of the original signal is performed, and the data set is subsetted according to the decomposition result to obtain a preliminary partitioning result; Based on the preliminary division results, recursive division is performed until the preset recursive end condition is reached to obtain the final division result; A decision tree is constructed based on the final division results, and the combustion temperature of the gas turbine is predicted using the decision tree.

2. The method for predicting combustion temperature of a gas turbine according to claim 1, wherein: The constructing of the envelope based on the data set includes: The envelope is constructed using the interpolation method. An upper envelope in the signal connects the maximum points of the interpolation points, and a lower envelope in the signal connects the minimum points of the interpolation points.

3. The method for predicting combustion temperature of a gas turbine according to claim 2, wherein: The step of performing a first judgment based on the envelope, terminating the first judgment if the judgment result is satisfied, and reconstructing the envelope if the judgment result is not satisfied until the judgment result is satisfied includes: The first-order intrinsic mode function is iteratively separated from the original signal of the data set. In each iteration, the signal is made to meet the conditions of the intrinsic mode function by calculating the average function and the difference. When the conditions of the intrinsic mode function are met, the iteration is stopped; otherwise, the current difference signal is used as the new original signal and the iteration is continued until the conditions of the intrinsic mode function are met.

4. The method for predicting combustion temperature of a gas turbine according to claim 3, wherein: The data set is divided into subsets according to the decomposition result to obtain preliminary division results including: Calculate the mean square error of the EMD decomposition result of the original signal. If the mean square error is greater than the set value, it is classified as abnormal data, otherwise it is normal data.

5. The method for predicting combustion temperature of a gas turbine according to claim 4, wherein: Based on the preliminary division result, recursive division is performed until a preset recursive end condition is reached to obtain a final division result; Building a decision tree based on the final partitioning results includes: Recursive partitioning: recursive partitioning is performed based on the preliminary partitioning results. Each partitioning determines the output of the partitioning unit and adds a new node. The feature space is divided by a heuristic method. Each time, all the values ​​of all the features in the current set are examined, and the optimal split point is selected according to the square error minimization criterion.

6. The method for predicting combustion temperature of a gas turbine according to claim 5, wherein: Based on the preliminary division result, recursive division is performed until a preset recursive end condition is reached to obtain a final division result; Building a decision tree based on the final partitioning results also includes: Select the split point: examine the features and their values ​​one by one, traverse the feature variables, scan the split points for the fixed features, and select the feature and split point combination that minimizes the square error; Divide the region: Divide the region by combining the selected features and the split points, and determine the output value of the corresponding region; Repeated division: Repeat the steps of selecting the split point and dividing the area for the two sub-areas until the preset stop condition is met; The input space is divided into multiple regions, and finally a decision tree is generated, where each region corresponds to a leaf node output value.

7. The method for predicting combustion temperature of a gas turbine according to claim 6, wherein: The method of using a decision tree to predict the combustion temperature of a gas turbine includes: The input sample is traversed along the path of the decision tree until it reaches a leaf node, and the output value of the leaf node is used as the prediction result.

8. A gas turbine combustion temperature prediction system, applying the method according to any one of claims 1 to 7, characterized in that: include: An envelope curve construction module is used to obtain combustion temperature data to construct a data set, and to construct an envelope curve based on the data set; A first judgment module is configured to perform a first judgment based on the envelope, terminate the first judgment if the judgment result is satisfied, and reconstruct the envelope if the judgment result is not satisfied until the judgment result is satisfied; A preliminary partitioning module is used to perform EMD decomposition of the original signal based on the envelope after the first judgment is terminated, and to perform subset partitioning on the data set according to the decomposition result to obtain a preliminary partitioning result; The decision tree prediction module is used to perform recursive partitioning based on the preliminary partitioning results until the preset recursive end condition is reached to obtain the final partitioning result; A decision tree is constructed based on the final division results, and the combustion temperature of the gas turbine is predicted using the decision tree.

9. A computing device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images