Gas turbine oscillation combustion fault detection method and system

Abstract

The invention discloses a gas turbine oscillation combustion fault detection method and system, and relates to the technical field of gas turbine fault detection, and the method comprises the steps: collecting the data of a gas turbine for preprocessing; performing multi-scale analysis on the gas turbine data by using continuous wavelet transform, and analyzing and extracting a wavelet coefficient; classifying and distinguishing fault features and stable features; and taking the classified wavelet coefficient as the input of the convolutional neural network, and training the model until the diagnosis effect is optimal. According to the method, time-frequency analysis is carried out on flame pressure pulsation data by adopting continuous wavelet transform, wavelet coefficients are extracted, and data dimension reduction and feature extraction are realized. A convolutional neural network deep learning algorithm is used for flame feature classification, and compared with a traditional fault diagnosis method, the fault diagnosis time is shortened, the fault diagnosis speed is increased, and the calculation cost is reduced. The oscillation combustion fault detection algorithm provided by the invention has higher detection speed and higher accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of gas turbine fault detection, and in particular to a gas turbine oscillation combustion fault detection method and system. Background Art

[0002] Lean premixed swirl combustion is a combustion organization method commonly used in gas turbines. This combustion method can control the combustion temperature below 1650℃ and reduce NOx emissions. However, due to the proximity to the flameout limit, lean premixed combustion is prone to cause fluctuations in the heat release rate, which in turn causes pressure pulsations inside the combustion chamber, causing flow field disturbances, and further stimulating heat release rate fluctuations. This coupling process is the main factor in the occurrence of oscillating combustion failures in gas turbine combustion chambers. Once oscillating combustion failures are induced, they will seriously affect the safe operation of the entire unit and even cause catastrophic accidents. Therefore, it is necessary to develop a gas turbine oscillating combustion failure detection method with high diagnostic accuracy and strong generalization ability, so as to perform shutdown or active control operations in the early stage of the failure in a timely manner to ensure the safety of personnel and property.

[0003] Accurate detection of oscillating combustion faults is crucial to reducing the potential safety hazards and economic losses caused by them. Many foreign research institutions and scholars have made great efforts to monitor the dynamics of unstable combustion. The commonly used diagnostic method in engineering is to perform time-frequency analysis on pressure signals. However, when the amount of data is large, the existing data analysis method cannot guarantee the accuracy and speed of diagnosis, which may cause false alarms, alarm delays, etc., resulting in huge property losses.

[0004] As a supervised deep learning algorithm, Convolutional Neural Networks (CNN) has been widely used in various fields and has achieved convincing results, especially for computer vision tasks with the goal of feature extraction and classification. CNN has made great progress in sensor fault diagnosis, target detection, classification, noise suppression and change detection. In practical applications, CNN performs convolution in two spatial directions (x, y) to discover potential features hidden in the data.

[0005] The present invention proposes to detect the characteristics of combustion by applying CWT to pressure sensor signals or acceleration sensor signals, and combining it with artificial intelligence such as convolutional neural networks. Compared with advanced methods such as TVE, the advantage of this method is that there is no need to calibrate the threshold for each measured engine. The algorithm can automatically learn dynamic thresholds by simply looking at samples of pressure trajectories of similar combustion chambers, and can handle different engines and configurations. As the database expands, the generalization performance of the algorithm is expected to improve, thereby reducing the tendency of neural networks to overfit the training data. Summary of the invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by the present invention is: how to improve the accuracy and real-time performance of gas turbine oscillation combustion fault detection, especially to accurately classify and judge the combustion state under complex working conditions.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for detecting oscillating combustion faults of a gas turbine, comprising the following steps:

[0009] Collect gas turbine data and perform data preprocessing.

[0010] Based on the preprocessing results, the gas turbine data is analyzed at multiple scales using continuous wavelet transform, and synchronous analysis is performed in the time domain to extract the wavelet coefficients.

[0011] The combustion state is classified according to the obtained wavelet coefficients to distinguish between fault characteristics and stable characteristics.

[0012] The classified wavelet coefficients are used as the input of the convolutional neural network, the model is trained, and the accuracy is output.

[0013] Adjust the transformation model parameters according to the output results until the diagnostic effect is optimal.

[0014] As a preferred solution of the gas turbine oscillation combustion fault detection method described in the present invention, wherein: the data of the gas turbine is collected by using a sensor to collect the temperature, combustion efficiency and pressure data in the combustion chamber of the gas turbine;

[0015] The preprocessing is to use a low-pass filter to denoise the combustion chamber temperature, combustion efficiency and combustion chamber pressure data collected by the gas turbine, filter out the high-frequency noise in each signal, normalize the denoised temperature, combustion efficiency and pressure data, use the 3σ principle to identify and process the normalized data for outliers, calculate the mean and standard deviation of each signal, and identify outliers that exceed the range of three times the standard deviation, perform interpolation processing on the data of the outliers, align the time series of data from different sensors, fill in the missing data that occurs during the collection or transmission process through the interpolation method for the data after the time series alignment, segment the signal data after the time series alignment and data filling, and use the sliding window technology to divide the continuous data stream into time windows of fixed length.

[0016] As a preferred solution of the gas turbine oscillation combustion fault detection method described in the present invention, wherein: the continuous wavelet transform is used to perform multi-scale analysis on the gas turbine data, and synchronous analysis is performed in the time domain, and the wavelet coefficients are extracted by using the adaptive wavelet transform method to perform multi-scale analysis on the temperature, combustion efficiency and pressure signals in each time window after the time series alignment is completed;

[0017] The spectrum characteristics of temperature, combustion efficiency and pressure signals are analyzed, and the mother wavelet function is adaptively selected to decompose the signal through wavelet transform. The expression is:

[0018]

[0019] Among them, T c (x), E f (x), P c (x) are the combustion chamber temperature signal, combustion efficiency signal and pressure signal at time x, respectively, W T (a T ,b T )、W E (a E ,b E )、W P (a P ,b P ) are the wavelet transform results of the combustion chamber temperature signal, combustion efficiency signal and pressure signal, respectively. T 、a E 、a P are the wavelet transform scale factors of the combustion chamber temperature signal, combustion efficiency signal and pressure signal, respectively, and b T 、b E 、b P They are the wavelet transform displacement factors of the combustion chamber temperature signal, combustion efficiency signal and pressure signal, It is the form after scaling and translation of the mother wavelet function.

[0020] As a preferred solution of the gas turbine oscillation combustion fault detection method described in the present invention, the extraction of wavelet coefficients also includes determining the contribution of the current signal at the time node according to the wavelet transform results of the combustion chamber temperature signal, the combustion efficiency signal and the pressure signal, and the expression is:

[0021] E T =|W T (a T ,b T )|

[0022] E E =|W E (a E ,bE )|

[0023] E P =|W P (a P ,b P )|

[0024]

[0025] Among them, E T 、E E 、E P They are the amplitudes of the combustion chamber temperature signal, combustion efficiency signal and pressure signal respectively.

[0026] As a preferred solution of the gas turbine oscillation combustion fault detection method described in the present invention, the extraction of wavelet coefficients also includes weighted calculation of the wavelet transform results of the combustion chamber temperature signal, the combustion efficiency signal and the pressure signal to output a comprehensive value, the expression is:

[0027]

[0028] in, are the weights of the combustion chamber temperature signal, combustion efficiency signal and pressure signal respectively.

[0029] As a preferred solution of the gas turbine oscillation combustion fault detection method described in the present invention, wherein: the classified wavelet coefficients are used as the input of the convolutional neural network to train the model, and the output accuracy includes when the comprehensive value is greater than the transition threshold, it is determined that the current combustion state is unstable, and when the comprehensive value is less than or equal to the transition threshold, it is determined that the current combustion state is stable;

[0030] The wavelet transform results and the classified stable and unstable data are used as the input of the convolutional neural network. The convolution layer performs convolution operations on the input wavelet coefficients through the convolution kernel to extract local features. In the fully connected layer, the local features extracted by the convolution layer are integrated and further processed to form a complete feature vector. The fully connected layer combines the features with the classification labels and outputs the prediction results of the combustion state. During the model training process, the weights and parameters of the convolutional neural network are continuously optimized by comparing the error between the output of the convolutional neural network and the actual classification label to complete the prediction of the combustion state.

[0031] As a preferred solution of a gas turbine oscillation combustion fault detection method described in the present invention, the continuous optimization of the weights and parameters of the convolutional neural network is to define a cross entropy loss function, calculate the error between the model prediction result and the actual classification label, and use the back propagation algorithm to calculate the gradient of the loss function relative to each network parameter layer by layer starting from the output layer, and propagate it to each layer of the convolutional neural network, update the weight and bias parameters, use the gradient descent algorithm, adjust the weight and bias parameters of the network according to the calculated gradient, and set the learning rate to control the step size of each parameter update, gradually reduce the loss value of the model through iteration, and use the validation set to monitor the model performance. When the loss value of the model reaches the preset convergence condition in the training set and the validation set in multiple consecutive iterations, the optimization process of the model parameters ends.

[0032] Another object of the present invention is to provide a gas turbine oscillation combustion fault detection system, which can effectively identify the stability of the combustion state by performing wavelet transform, classification and deep learning on the combustion chamber temperature, combustion efficiency and pressure signals, thereby solving the problems of insufficient accuracy, poor real-time performance and limited complex signal processing capabilities of the fault detection methods in the prior art.

[0033] To solve the above technical problems, the present invention provides the following technical solutions: a gas turbine oscillation combustion fault detection system, comprising: a data collection and preprocessing module, a wavelet transform module, a classification and judgment module, a convolutional neural network training module and a combustion state prediction and output module.

[0034] The data collection and preprocessing module uses sensors to collect temperature, combustion efficiency and pressure data in the combustion chamber of the gas turbine, denoises and normalizes the original signal data of the gas turbine, and aligns the time series.

[0035] The wavelet transform module performs wavelet transform on the preprocessed signal to generate wavelet coefficients.

[0036] The classification and judgment module is based on the wavelet coefficients generated by the wavelet transformation module, and classifies the combustion state through a set threshold value, and classifies the combustion state into stable or unstable.

[0037] The convolutional neural network training module trains the convolutional neural network by receiving classified wavelet coefficients and classification labels. The convolution layer extracts local features from the wavelet coefficients and performs deep learning in combination with the classification labels. The weights and parameters of the convolutional neural network are optimized through iterative training.

[0038] The combustion state prediction and output module uses a trained and optimized convolutional neural network to process input wavelet coefficients and classification labels in real time to predict the current combustion state.

[0039] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned gas turbine oscillation combustion fault detection method when executing the computer program.

[0040] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a gas turbine oscillation combustion fault detection method as described above.

[0041] Beneficial effects of the present invention: The present invention uses continuous wavelet transform to perform time-frequency analysis on combustion chamber temperature, combustion efficiency and pressure signals, extracts wavelet coefficients, and realizes data dimension reduction and feature extraction. The convolutional neural network deep learning algorithm is used to classify flame features. Compared with traditional fault diagnosis methods, it shortens the fault diagnosis time, improves the fault diagnosis speed, and greatly reduces the computing cost. The convolutional neural network improves the accuracy of fault diagnosis due to its self-supervised learning algorithm. Compared with the prior art, the oscillation combustion fault detection algorithm proposed by the present invention has a faster detection speed and higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing 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 creative work. Among them:

[0043] Figure 1 An overall flow chart of a gas turbine oscillation combustion fault detection method provided by the first embodiment of the present invention;

[0044] Figure 2 A weighted comprehensive numerical continuous wavelet transform spectrum diagram of combustion chamber temperature, combustion efficiency and pressure signals with an oscillation frequency of 180 Hz in a gas turbine oscillation combustion fault detection method provided in a third embodiment of the present invention;

[0045] Figure 3 A weighted comprehensive numerical continuous wavelet transform spectrum diagram of combustion chamber temperature, combustion efficiency and pressure signals under stable working conditions in a gas turbine oscillation combustion fault detection method provided by the third embodiment of the present invention;

[0046] Figure 4 The accuracy of the diagnosis result using the original combustion chamber temperature, combustion efficiency and pressure signal weighted comprehensive value as the CNN network in a gas turbine oscillation combustion fault detection method provided by the third embodiment of the present invention;

[0047] Figure 5The third embodiment of the present invention provides a method for detecting oscillating combustion faults in a gas turbine, in which the wavelet coefficients of the weighted comprehensive numerical transformation of the combustion chamber temperature, combustion efficiency and pressure signals are used as the accuracy of the result of CNN neural network diagnosis. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0049] Example 1, reference Figure 1 According to an embodiment of the present invention, a method for detecting an oscillating combustion fault of a gas turbine is provided, comprising:

[0050] Collect gas turbine data and perform data preprocessing;

[0051] Based on the preprocessing results, the gas turbine data is analyzed at multiple scales using continuous wavelet transform, and synchronous analysis is performed in the time domain to extract wavelet coefficients.

[0052] The combustion state is classified according to the obtained wavelet coefficients to distinguish between fault characteristics and stable characteristics;

[0053] The classified wavelet coefficients are used as the input of the convolutional neural network, the model is trained, and the accuracy is output;

[0054] Adjust the transformation model parameters according to the output results until the diagnostic effect is optimal.

[0055] The gas turbine data collection includes using sensors to collect temperature, combustion efficiency and pressure data in the gas turbine combustion chamber;

[0056] The preprocessing is to use a low-pass filter to denoise the combustion chamber temperature, combustion efficiency and combustion chamber pressure data collected by the gas turbine, filter out the high-frequency noise in each signal, normalize the denoised temperature, combustion efficiency and pressure data, use the 3σ principle to identify and process the normalized data for outliers, calculate the mean and standard deviation of each signal, and identify outliers that exceed the range of three times the standard deviation, perform interpolation processing on the data of the outliers, align the time series of data from different sensors, fill in the missing data that occurs during the collection or transmission process through the interpolation method for the data after the time series alignment, segment the signal data after the time series alignment and data filling, and use the sliding window technology to divide the continuous data stream into time windows of fixed length.

[0057] After the time series alignment is completed, the temperature, combustion efficiency and pressure signals are analyzed on a multi-scale basis using the adaptive wavelet transform method in each time window;

[0058] The spectrum characteristics of temperature, combustion efficiency and pressure signals are analyzed, the mother wavelet function is adaptively selected and the signal is decomposed by wavelet transform. The expression is:

[0059]

[0060] Among them, T c (x), E f (x), P c (x) are the combustion chamber temperature signal, combustion efficiency signal and pressure signal at time x, respectively, W T (a T ,b T )、W E (a E ,b E )、W P (a P ,b P ) are the wavelet transform results of the combustion chamber temperature signal, combustion efficiency signal and pressure signal, respectively. T 、a E 、a P are the wavelet transform scale factors of the combustion chamber temperature signal, combustion efficiency signal and pressure signal, respectively, and b T 、b E 、b P They are the wavelet transform displacement factors of the combustion chamber temperature signal, combustion efficiency signal and pressure signal,

[0061] It is the form after scaling and translation of the mother wavelet function.

[0062] The contribution of the current signal at the time node is determined based on the wavelet transform results of the combustion chamber temperature signal, combustion efficiency signal and pressure signal. The expression is:

[0063] E T =|W T (a T ,b T )|

[0064] E E =|W E (a E ,b E )|

[0065] E P =|W P (aP ,b P )|

[0066]

[0067] Among them, E T 、E E 、E P They are the amplitudes of the combustion chamber temperature signal, combustion efficiency signal and pressure signal respectively.

[0068] The wavelet transform results of the combustion chamber temperature signal, combustion efficiency signal and pressure signal are weighted and calculated to output a comprehensive value, the expression is:

[0069]

[0070] in, are the weights of the combustion chamber temperature signal, combustion efficiency signal and pressure signal respectively.

[0071] When the comprehensive value is greater than the transition threshold, it is determined that the current combustion state is unstable; when the comprehensive value is less than or equal to the transition threshold, it is determined that the current combustion state is stable;

[0072] The wavelet transform results and the classified stable and unstable data are used as the input of the convolutional neural network. The convolution layer performs convolution operations on the input wavelet coefficients through the convolution kernel to extract local features. In the fully connected layer, the local features extracted by the convolution layer are integrated and further processed to form a complete feature vector. The fully connected layer combines the features with the classification labels and outputs the prediction results of the combustion state. During the model training process, the weights and parameters of the convolutional neural network are continuously optimized by comparing the error between the output of the convolutional neural network and the actual classification label to complete the prediction of the combustion state.

[0073] It should be further explained that the local features in the present invention refer to the signal features within a specific frequency range and at a specific time point extracted from these wavelet coefficients through convolution operations. These local features are the signal response at a specific moment, reflecting the change and characteristic pattern of the signal within a local range, and are key indicators for judging the combustion state.

[0074] It should be further explained that in the continuous optimization of the weights and parameters of the convolutional neural network, the cross entropy loss function is first defined, the error between the model prediction result and the actual classification label is calculated, and the gradient of the loss function relative to each network parameter is calculated layer by layer starting from the output layer using the back propagation algorithm, and propagated to each layer of the convolutional neural network, and the weights and bias parameters are updated. The gradient descent algorithm is used to adjust the weights and bias parameters of the network according to the calculated gradient, and the learning rate is set to control the step size of each parameter update. The loss value of the model is gradually reduced through iteration, and the validation set is used to monitor the model performance. When the loss value of the model reaches the preset convergence condition in the training set and the validation set in multiple consecutive iterations, the optimization process of the model parameters ends.

[0075] Embodiment 2 is an embodiment of the present invention, which provides a system for detecting gas turbine oscillation combustion faults, including: a data collection and preprocessing module, a wavelet transform module, a classification and judgment module, a convolutional neural network training module, and a combustion state prediction and output module.

[0076] The data collection and preprocessing module uses sensors to collect temperature, combustion efficiency and pressure data in the gas turbine combustion chamber, denoises and normalizes the original signal data of the gas turbine, and aligns the time series.

[0077] The wavelet transform module performs wavelet transform on the preprocessed signal to generate wavelet coefficients.

[0078] The classification and judgment module is based on the wavelet coefficients generated by the wavelet transform module. It classifies the combustion state through the set threshold and divides the combustion state into stable or unstable.

[0079] The convolutional neural network training module trains the convolutional neural network by receiving classified wavelet coefficients and classification labels. The convolution layer extracts local features from the wavelet coefficients and combines them with the classification labels for deep learning. Through iterative training, the weights and parameters of the convolutional neural network are optimized.

[0080] The combustion state prediction and output module uses a trained and optimized convolutional neural network to process the input wavelet coefficients and classification labels in real time to predict the current combustion state.

[0081] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0082] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0083] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0084] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0085] Example 3, reference Figure 2-Figure 5 This is another embodiment of the present invention. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments. This embodiment experiments the existing traditional method and the method of this embodiment respectively.

[0086] In order to verify the effectiveness of the method proposed in this paper, the present invention conducted an experimental study on a self-built swirl lean premixed gas turbine model burner. When the equivalence ratio was 0.63, the flame produced an oscillation phenomenon with a main frequency of 180Hz. The wavelet coefficients of the pressure pulsation data, temperature and combustion efficiency of the flame were extracted using the continuous wavelet transform time-frequency analysis method, and the wavelet coefficients were input into the CNN network for fault classification and identification.

[0087] from Figure 2 and Figure 3 It can be seen that when the flame is in a critical state of unstable combustion, the critical value of the wavelet coefficient is 0.02. After the flame enters the unstable combustion state, the wavelet coefficient is much larger than 0.02. When the original pressure signal is used as the input of the trained convolutional neural network, Figure 4 The maximum error of the diagnosis result is 5%. After the transformed wavelet coefficients are used as the input of the convolutional neural network, Figure 5 At this time, the maximum error of the diagnosis result is 2%.

[0088] The advantages of using wavelet coefficients for CNN network training in the present invention are:

[0089] Compared with the traditional method of directly using the pressure pulsation signal as the input of CNN, the continuous wavelet transform can filter and reduce the noise of the original pressure signal, improve the quality of the original pressure signal, and improve the readability and clarity of the signal by suppressing or filtering out the noise and interference in the signal, making it easier to analyze and process.

[0090] For the CNN diagnosis process, the pressure signal after wavelet transformation is smoothed, and bad points or bad values ​​caused by background noise, equipment accuracy drift or manual measurement errors are eliminated, thereby improving the accuracy of subsequent CNN network diagnosis and reducing errors.

[0091] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. 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 detecting oscillating combustion faults in a gas turbine, characterized in that: include: Collect gas turbine data and perform data preprocessing; Based on the preprocessing results, the gas turbine data is analyzed at multiple scales using continuous wavelet transform, and synchronous analysis is performed in the time domain to extract wavelet coefficients. The combustion state is classified according to the obtained wavelet coefficients to distinguish between fault characteristics and stable characteristics; The classified wavelet coefficients are used as the input of the convolutional neural network, the model is trained, and the accuracy is output; Adjust the transformation model parameters according to the output results until the diagnostic effect is optimal.

2. A gas turbine oscillation combustion fault detection method according to claim 1, characterized in that: The gas turbine data collection includes using sensors to collect temperature, combustion efficiency and pressure data in the gas turbine combustion chamber; The preprocessing is to use a low-pass filter to denoise the combustion chamber temperature, combustion efficiency and combustion chamber pressure data collected by the gas turbine, filter out the high-frequency noise in each signal, normalize the denoised temperature, combustion efficiency and pressure data, use the 3σ principle to identify and process the normalized data for outliers, calculate the mean and standard deviation of each signal, and identify outliers that exceed the range of three times the standard deviation, perform interpolation processing on the data of the outliers, align the time series of data from different sensors, fill in the missing data that occurs during the collection or transmission process through the interpolation method for the data after the time series alignment, segment the signal data after the time series alignment and data filling, and use the sliding window technology to divide the continuous data stream into time windows of fixed length.

3. A gas turbine oscillation combustion fault detection method according to claim 2, characterized in that: The method of using continuous wavelet transform to perform multi-scale analysis on gas turbine data and perform synchronous analysis in the time domain is to extract wavelet coefficients after the time series alignment is completed, and then perform multi-scale analysis on the temperature, combustion efficiency and pressure signals in each time window by using adaptive wavelet transform method; The spectrum characteristics of temperature, combustion efficiency and pressure signals are analyzed, the mother wavelet function is adaptively selected and the signal is decomposed by wavelet transform. The expression is: Among them, T c (x), E f (x), P c (x) are the combustion chamber temperature signal, combustion efficiency signal and pressure signal at time x, respectively, W T (a T ,b T )、W E (a E ,b E )、W P (a P ,b P ) are the wavelet transform results of the combustion chamber temperature signal, combustion efficiency signal and pressure signal, respectively. T 、a E 、a P are the wavelet transform scale factors of the combustion chamber temperature signal, combustion efficiency signal and pressure signal, respectively, and b T 、b E 、b P They are the wavelet transform displacement factors of the combustion chamber temperature signal, combustion efficiency signal and pressure signal, It is the form after scaling and translation of the mother wavelet function.

4. A gas turbine oscillation combustion fault detection method according to claim 3, characterized in that: The extraction of wavelet coefficients also includes determining the contribution of the current signal at the time node according to the wavelet transform results of the combustion chamber temperature signal, the combustion efficiency signal and the pressure signal, and the expression is: HAVE BEEN T =|W T (a T ,b T )| HAVE BEEN E =|W E (a E ,b E )| HAVE BEEN P =|W P (a P ,b P )| Among them, E T 、E E 、E P They are the amplitudes of the combustion chamber temperature signal, combustion efficiency signal and pressure signal respectively.

5. A gas turbine oscillation combustion fault detection method according to claim 4, characterized in that: The extraction of wavelet coefficients also includes weighted calculation of the wavelet transform results of the combustion chamber temperature signal, the combustion efficiency signal and the pressure signal to output a comprehensive value, the expression is: in, are the weights of the combustion chamber temperature signal, combustion efficiency signal and pressure signal respectively.

6. A gas turbine oscillation combustion fault detection method according to claim 5, characterized in that: The classified wavelet coefficients are used as the input of the convolutional neural network to train the model, and the output accuracy includes when the comprehensive value is greater than the transition threshold, it is determined that the current combustion state is unstable, and when the comprehensive value is less than or equal to the transition threshold, it is determined that the current combustion state is stable; The wavelet transform results and the classified stable and unstable data are used as the input of the convolutional neural network. The convolution layer performs convolution operations on the input wavelet coefficients through the convolution kernel to extract local features. In the fully connected layer, the local features extracted by the convolution layer are integrated and further processed to form a complete feature vector. The fully connected layer combines the features with the classification labels and outputs the prediction results of the combustion state. During the model training process, the weights and parameters of the convolutional neural network are continuously optimized by comparing the error between the output of the convolutional neural network and the actual classification label to complete the prediction of the combustion state.

7. A gas turbine oscillation combustion fault detection method according to claim 6, characterized in that: The continuous optimization of the weights and parameters of the convolutional neural network is to define a cross entropy loss function, calculate the error between the model prediction result and the actual classification label, use the back propagation algorithm, start from the output layer, calculate the gradient of the loss function relative to each network parameter layer by layer, and propagate it to each layer of the convolutional neural network, update the weight and bias parameters, use the gradient descent algorithm, adjust the weight and bias parameters of the network according to the calculated gradient, set the learning rate to control the step size of each parameter update, gradually reduce the loss value of the model through iteration, and use the validation set to monitor the model performance. When the loss value of the model reaches the preset convergence condition in the training set and the validation set in multiple consecutive iterations, the optimization process of the model parameters ends.

8. A system using a gas turbine oscillation combustion fault detection method according to any one of claims 1 to 7, characterized in that: It includes data collection and preprocessing module, wavelet transform module, classification and judgment module, convolutional neural network training module and combustion state prediction and output module; The data collection and preprocessing module uses sensors to collect temperature, combustion efficiency and pressure data in the combustion chamber of the gas turbine, performs denoising and normalization processing on the original signal data of the gas turbine, and aligns the time series; The wavelet transform module performs wavelet transform on the preprocessed signal to generate wavelet coefficients; The classification and judgment module is based on the wavelet coefficients generated by the wavelet transform module, and classifies the combustion state by a set threshold value, and divides the combustion state into stable or unstable. The convolutional neural network training module trains the convolutional neural network by receiving the classified wavelet coefficients and classification labels. The convolution layer extracts local features from the wavelet coefficients and performs deep learning in combination with the classification labels. The weights and parameters of the convolutional neural network are optimized through iterative training. The combustion state prediction and output module uses a trained and optimized convolutional neural network to process input wavelet coefficients and classification labels in real time to predict the current combustion state.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of a gas turbine oscillation combustion fault detection method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a gas turbine oscillating combustion fault detection method as claimed in any one of claims 1 to 7 are implemented.

Images