Combustion pulsation pressure trend monitoring method and early warning system

Abstract

The present application relates to the technical field of combustion process optimization, and discloses a combustion pulsation pressure trend monitoring method and an early warning system, including: obtaining combustion pulsation pressure data in a combustion chamber; predicting the pressure change trends of different frequency bands in the combustion chamber based on the combustion pulsation pressure data, and obtaining a multi-band pressure prediction trend; and issuing an early warning of pressure abnormalities based on the multi-band pressure prediction trend; the present application predicts the pressure change trend of each frequency band based on the pulsation pressure data of different frequency bands in the combustion chamber, and can capture small fluctuations and abnormal changes before a fault, discover potential pressure fluctuation problems in advance, and dynamically detect pressure abnormalities in different frequency bands, accurately identify the abnormal pressure location and abnormal type, and can issue an early warning of potential abnormalities, optimize the combustion process, accurately locate the source and nature of the problem, improve the maintenance efficiency of staff, and thus reduce the chance of accidents.

Description

Technical Field

[0001] The present invention relates to the technical field of combustion process optimization, and in particular to a combustion pulsation pressure trend monitoring method and an early warning system. Background Art

[0002] Combustion pulsation pressure is a key physical phenomenon generated during the combustion process. It reflects the interaction of multiple factors within the combustion chamber, including flame propagation, fuel-air mixing, and the dynamic response of the combustion chamber structure. The magnitude and frequency of combustion pulsation pressure have a significant impact on the operational stability, safety, and efficiency of the equipment. Traditional combustion pulsation pressure monitoring methods rely primarily on single or multiple pressure sensors to measure the pressure at specific locations within the combustion chamber. This makes it difficult to fully reflect the pressure distribution within the entire combustion chamber. These methods are unable to conduct in-depth analysis of the frequency characteristics of combustion pulsation pressure and ignore the differences in abnormal pressure thresholds across different frequency bands. Furthermore, these methods only analyze real-time pressure anomalies and fail to proactively prevent potential anomalies.

[0003] For example, a Chinese patent with authorization announcement number CN110966100B discloses a combustion oscillation monitoring device and method, which include: collecting multiple dynamic signals from a combustion chamber and its periphery, the multiple dynamic signals including at least the pressure of the combustion chamber; analyzing each dynamic signal in the time domain, comparing the time domain analysis result of each dynamic signal with the corresponding first threshold, and judging whether the time domain analysis results of all dynamic signals are lower than the corresponding first threshold, if so, entering the early combustion oscillation diagnosis step, otherwise entering the next step; analyzing each dynamic signal in the frequency domain to obtain an indication value of each dynamic signal; fusing the indication values ​​to obtain a total indication value; judging whether the total indication value is lower than a second threshold, if so, judging that no combustion oscillation has occurred, otherwise judging that combustion oscillation has occurred.

[0004] The above-mentioned prior art has the problems raised by this background technology: since the combustion process in the combustion chamber is complex and changeable, it is difficult for a single sensor to fully reflect the pressure distribution in the entire combustion chamber; it is impossible to conduct in-depth analysis of the frequency characteristics of the combustion pulsation pressure, and ignores the differences in abnormal pressure thresholds in different frequency bands; it only analyzes real-time pressure abnormalities, and cannot prevent potential abnormalities in advance; in order to solve at least one of the above problems, the present invention proposes a combustion pulsation pressure trend monitoring method and early warning system. Summary of the Invention

[0005] In response to the shortcomings of the prior art, the main purpose of the present invention is to provide a combustion pulsation pressure trend monitoring method and early warning system that can effectively solve the problems in the background art. The specific technical solutions of the present invention are as follows:

[0006] Combustion pulsation pressure trend monitoring method, including:

[0007] Obtain combustion pulsation pressure data in the combustion chamber;

[0008] According to the combustion pulsation pressure data, the pressure variation trends of different frequency bands in the combustion chamber are predicted respectively to obtain a multi-band pressure prediction trend;

[0009] According to the multi-band pressure prediction trend, an early warning of abnormal pressure conditions is issued.

[0010] Specifically, the pressure change trends of different frequency bands in the combustion chamber are predicted based on the combustion pulsation pressure data to obtain a multi-band pressure prediction trend, including:

[0011] Perform frequency domain conversion on the combustion pulsation pressure data to obtain frequency domain pressure data;

[0012] Dividing the frequency domain pressure data into a plurality of frequency bands according to a preset frequency band range, wherein the plurality of frequency bands include a high frequency band and a non-high frequency band;

[0013] The pressure change trend of each frequency band is predicted separately to obtain the multi-band pressure prediction trend.

[0014] Specifically, predicting the pressure change trend of each frequency band to obtain the multi-band pressure prediction trend includes:

[0015] For high-frequency pressure data, a preset first pressure prediction model is used to predict the pressure change trend to obtain a first pressure change trend;

[0016] For non-high-frequency band pressure data, a pressure change trend prediction is performed according to a preset second pressure prediction model to obtain a second pressure change trend.

[0017] Specifically, for the high-frequency pressure data, using a preset first pressure prediction model to predict the pressure change trend to obtain the first pressure change trend includes:

[0018] According to the preset high-frequency fault types, multiple target frequency spectra are obtained, wherein the high-frequency fault types include incomplete combustion, excessive thermoacoustic oscillation, local cracks in the flame tube, and abnormal turbulence characteristics;

[0019] dividing the high-frequency band pressure data according to the multiple target spectra to obtain multi-spectrum pressure data;

[0020] Based on the historical high-frequency pressure data, feature extraction is performed on each spectrum pressure data to obtain a multi-spectrum pressure feature set;

[0021] According to the multi-spectral pressure feature set, each spectrum pressure is predicted respectively using a preset first pressure prediction model to obtain a first pressure change trend.

[0022] Specifically, for the non-high-frequency band pressure data, performing pressure change trend prediction according to a preset second pressure prediction model to obtain a second pressure change trend includes:

[0023] Feature extraction is performed based on historical non-high frequency band pressure data to obtain a non-high frequency band pressure feature set;

[0024] According to the non-high-frequency band pressure feature set, the pressure is predicted using a preset second pressure prediction model to obtain a second pressure change trend.

[0025] Specifically, the step of issuing an early warning for abnormal pressure conditions based on the multi-band pressure prediction trend includes:

[0026] For the first pressure change trend in the high-frequency band, a preset abnormal warning model is used to issue an early warning of pressure anomalies;

[0027] For the second pressure change trend in the non-high frequency band, an early warning of abnormal pressure conditions is issued based on the preset early warning value and high alarm value.

[0028] Specifically, the first pressure change trend in the high frequency band is used to issue an early warning of pressure anomaly using a preset abnormality early warning model, including:

[0029] According to the first pressure change trend, a change trend curve of each target spectrum pressure is drawn;

[0030] According to the change trend curve, the slope and amplitude increment of the curve are calculated to obtain the trend characteristics of each target spectrum pressure;

[0031] According to the preset high-frequency fault type and historical fault pressure data, the alarm threshold is updated using the preset fault spectrum model to obtain the updated alarm threshold for each spectrum pressure;

[0032] Comparing the trend feature with the updated alarm threshold of the corresponding spectrum to obtain the pressure anomaly location and anomaly type;

[0033] According to the abnormal pressure location and abnormal type, an early warning is issued through a preset abnormal warning model.

[0034] Specifically, obtaining the combustion pulsation pressure data in the combustion chamber includes:

[0035] Performing windowing processing on the combustion pulsation pressure data to obtain windowed pressure data;

[0036] resampling the windowed pressure data to obtain resampled data;

[0037] The resampled data are combined to obtain updated combustion pulsation pressure data.

[0038] Specifically, resampling the windowed pressure data to obtain resampled data includes:

[0039] Extracting the windowed pressure data to obtain extracted data;

[0040] Interpolation processing is performed on the extracted data to obtain resampled data.

[0041] The combustion pulsation pressure early warning system is used to implement the combustion pulsation pressure trend monitoring method, including:

[0042] A data acquisition module, which acquires combustion pulsation pressure data in the combustion chamber;

[0043] A pressure change trend prediction module predicts the pressure change trends of different frequency bands in the combustion chamber based on the combustion pulsation pressure data to obtain a multi-band pressure prediction trend;

[0044] The early warning module issues an early warning of abnormal pressure conditions based on the multi-band pressure prediction trend.

[0045] Compared with the prior art, this application has the following beneficial effects:

[0046] This application is based on the pulsating pressure data of different frequency bands in the combustion chamber, and predicts the pressure change trend of each frequency band respectively, which can detect potential pressure fluctuation problems in advance, and dynamically detect pressure anomalies in different frequency bands. It can accurately identify the abnormal pressure location and abnormal type, and can provide early warning for potential anomalies, optimize the combustion process, achieve full combustion of fuel and a stable combustion state, accurately locate the source and nature of the problem, avoid the smooth progress of the combustion process due to abnormal conditions in the combustion chamber, improve the maintenance efficiency and abnormal response speed of staff, and enhance the supervision of the combustion system, thereby reducing the chance of accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flowchart of the combustion pulsation pressure trend monitoring method in Example 1 of the present invention;

[0048] Figure 2 This is a flowchart of the high-frequency band pressure anomaly detection process in Example 1 of the present invention;

[0049] Figure 3 This is a flowchart of the high-frequency band pressure anomaly warning process in Example 1 of the present invention;

[0050] Figure 4This is a structural diagram of the combustion pulsation pressure warning system in Example 3 of the present invention. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0052] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0054] Example 1:

[0055] This embodiment provides a method for monitoring combustion pulsation pressure trends, such as Figure 1 As shown, the combustion pulsation pressure trend monitoring method includes:

[0056] S101, obtaining combustion pulsation pressure data in the combustion chamber;

[0057] S102, predicting pressure change trends in different frequency bands within the combustion chamber based on the combustion pulsation pressure data to obtain a multi-band pressure prediction trend;

[0058] S103: issuing an early warning of abnormal pressure conditions based on the multi-band pressure prediction trend.

[0059] Conventional systems for identifying and warning abnormal pressure conditions within the combustion chamber based on real-time fluctuating pressure data and static thresholds cannot meet the requirements for early warning and complex operating conditions. The present embodiment uses fluctuating pressure data to perform long-term predictions of pressure trends within the combustion chamber, enabling early detection of potential abnormal pressure conditions within the combustion chamber, providing early warning and facilitating the development of timely maintenance plans.

[0060] Specifically, such as Figure 1, obtain the pulsating pressure data of the continuous operation of the gas turbine, perform data preprocessing on it to prepare for subsequent analysis, perform spectrum analysis on the preprocessed data, analyze the trend of each frequency band respectively, input the analysis results into the trained LSTM model, and at the same time combine the real fault spectrum data and the spectrum data of expert experience, references, and similar equipment to set the fault judgment threshold, and dynamically adjust the threshold to realize fault judgment; if the judgment result is correct, the process ends; when the judgment result is inaccurate, the LSTM model is retrained and then judged again until it is accurate. When a fault is predicted, a fault warning is issued.

[0061] Specifically, high-temperature pulsating pressure sensors are installed in the combustion chamber. These sensors can withstand the high temperature environment within the combustion chamber and capture the pulsating pressure signals generated during the combustion process in real time. This pulsating pressure data is collected in real time during the combustion process to reflect the state of the combustion process. Based on the combustion pulsating pressure data, the pressure variation trends in different frequency bands within the combustion chamber are predicted, and the long-term pressure variation trend of each frequency band is obtained. Each frequency band reflects different pressure characteristics. By predicting the pressure variation trends of different frequency bands, the dynamic characteristics of the pressure in the combustion chamber can be fully understood, potential pressure fluctuation problems can be identified in advance, and a more accurate basis for combustion control and optimization can be provided.

[0062] Specifically, different pressure anomaly analysis methods are used to perform an anomaly analysis on the pressure change trends in different frequency bands, which can specifically detect the specific pressure anomaly location and anomaly type, realize dynamic anomaly detection, and dynamically update the alarm threshold of each frequency band according to historical pressure data and fault type. When the predicted pressure trend exceeds the alarm threshold, an early warning is issued to send an early warning signal to relevant personnel and systems, reminding them to take timely measures to deal with it. By using dynamic thresholds for abnormal warnings, the alarm threshold can be updated in real time according to the combustion conditions in the combustion chamber, thereby improving the system's fault identification ability and the accuracy of abnormal situation detection, and can more accurately locate the source and nature of the problem, thereby improving the maintenance efficiency and abnormal response speed of the staff.

[0063] When abnormal pressure is detected in the combustion chamber, the system will alert the operator through a preset alarm mechanism. The alarm mechanism can warn the operator through sound, light or other means to remind them to pay attention to the abnormal situation in the combustion chamber. At the same time, the system can also record abnormal data and related information for subsequent analysis and processing. By promptly reminding the operator to pay attention to the abnormal situation in the combustion chamber, potential safety risks can be avoided, the supervision of the combustion system can be improved, and the chance of accidents can be reduced. According to the predicted state of the combustion chamber, the staff adjusts the combustion process by controlling the combustion speed and temperature. From fine adjustment of the combustion conditions to stable control of the overall combustion process, the comprehensive optimization of the combustion process ensures the stability of the production process and provides a solid guarantee for the consistency and reliability of product quality.

[0064] This application is based on the pulsating pressure data of different frequency bands in the combustion chamber, and predicts the pressure change trend of each frequency band respectively, which can detect potential pressure fluctuation problems in advance, and dynamically detect pressure anomalies in different frequency bands. It can accurately identify the abnormal pressure location and abnormal type, and can provide early warning for potential anomalies, optimize the combustion process, achieve full combustion of fuel and a stable combustion state, accurately locate the source and nature of the problem, avoid the smooth progress of the combustion process due to abnormal conditions in the combustion chamber, improve the maintenance efficiency and abnormal response speed of staff, and enhance the supervision of the combustion system, thereby reducing the chance of accidents.

[0065] Furthermore, the pressure change trends of different frequency bands in the combustion chamber are predicted based on the combustion pulsation pressure data to obtain a multi-band pressure prediction trend, including:

[0066] S201, performing frequency domain conversion on the combustion pulsation pressure data to obtain frequency domain pressure data;

[0067] S202, dividing the frequency domain pressure data into a plurality of frequency bands according to a preset frequency band range, wherein the plurality of frequency bands include a high frequency band and a non-high frequency band;

[0068] S203 , predicting the pressure change trend of each frequency band respectively to obtain a multi-frequency band pressure prediction trend.

[0069] In this embodiment, the resampled combustion pulsation pressure data is converted into frequency domain data using Fourier transform. In the combustion pulsation pressure analysis, the time-varying pressure signal is decomposed into the sum of sinusoidal waves of different frequencies using Fourier transform, thereby revealing the frequency components in the signal. Frequency domain analysis can intuitively display the frequency components in the signal. Frequency band division is based on the frequency components of the signal. In the combustion pulsation pressure analysis, the signal can be divided into a high-frequency band and a non-high-frequency band based on the frequency range. This division facilitates the detection of pressure anomalies within different frequency ranges. Specifically, the non-high-frequency band generally refers to a frequency band below a certain threshold, for example, a frequency band below 2 kHz is considered a non-high-frequency band, and the high-frequency band generally refers to a frequency band above a certain threshold, for example, a frequency band above 2 kHz is considered a high-frequency band. Frequency band division facilitates targeted detection of pressure anomalies within different frequency ranges, thereby improving the accuracy and efficiency of pressure anomaly detection.

[0070] Specifically, the pressure data within each frequency band has its own unique variation patterns and characteristics. Predicting the pressure variation trends in different frequency bands separately, and using the trends and patterns in historical pressure data to predict pressure variation trends in future time periods, can provide richer and more accurate information for the combustion system's operating status assessment, fault diagnosis, and optimized control, and enable a more comprehensive and detailed understanding of the dynamic changes in pressure within the combustion chamber. For example, predicting pressure changes in high-frequency bands can detect abnormal fluctuations in combustion chemical reactions in advance; predicting pressure trends in non-high-frequency bands can understand the stability of the overall flow and fuel supply in the combustion chamber, providing a strong guarantee for the safe and stable operation of the combustion system.

[0071] Furthermore, the pressure change trend of each frequency band is predicted respectively to obtain a multi-band pressure prediction trend, including:

[0072] S301: For high-frequency pressure data, use a preset first pressure prediction model to predict a pressure change trend to obtain a first pressure change trend;

[0073] S302 : For non-high-frequency band pressure data, predict the pressure change trend according to a preset second pressure prediction model to obtain a second pressure change trend.

[0074] In this embodiment, during the combustion process in the combustion chamber, high-frequency pressure data is related to some rapidly changing physical phenomena and chemical reactions, such as local turbulence in the combustion process, micro-scale oscillations of the flame, and rapid changes in chemical reaction rates, etc., and non-high-frequency pressure data is related to some macroscopic phenomena and slow-changing processes in the combustion system, such as the overall airflow movement in the combustion chamber, low-frequency fluctuations in the fuel supply system, and stability of the burner. The pressure change trends of the high-frequency and non-high-frequency bands are predicted respectively, and the pressure prediction model is trained according to the historical pressure data of the frequency band to obtain a pre-trained pressure prediction model, and the pressure prediction model is used to predict the pressure change trend.

[0075] Furthermore, for the high-frequency pressure data, using a preset first pressure prediction model to predict the pressure change trend to obtain the first pressure change trend includes:

[0076] S401. Obtain multiple target frequency spectra based on preset high-frequency fault types, wherein the high-frequency fault types include incomplete combustion, excessive thermoacoustic oscillation, local cracks in the flame tube, and abnormal turbulence characteristics;

[0077] S402, dividing the high-frequency pressure data according to the multiple target spectra to obtain multi-spectrum pressure data;

[0078] S403, based on the historical high-frequency pressure data, extracting features from each spectrum pressure data to obtain a multi-spectrum pressure feature set;

[0079] S404 : Predict each spectral pressure separately using a preset first pressure prediction model according to the multi-spectral pressure feature set to obtain a first pressure change trend.

[0080] In this embodiment, different high-frequency fault types will have specific manifestations in the spectrum of combustion pulsation pressure. The physical process and phenomenon corresponding to each fault will cause pressure fluctuations in different frequency ranges. For example, for a fault of excessive thermal acoustic oscillation, the pressure data will show a rapid increase in amplitude at the 50th and 75th frequencies. Figure 2 Initial fault alarm thresholds are set based on references, experience with similar equipment, and expert knowledge. FFT spectrum analysis is performed on a large amount of high-frequency fault data at intervals of 1 second or 10 minutes. This data is used to obtain high-frequency spectrum data at different time points (T1, T2, ..., TN), such as 40x and 50x frequency bands. Spectral trends in each frequency band are analyzed and fed into a pre-trained LSTM recurrent neural network. The network performs operations through input, forget, and output gates, and is optimized using a loss function. Based on the training results, the network configuration is adjusted to obtain a fault prediction model, enabling the monitoring of high-frequency faults.

[0081] Specifically, the pressure data in the high-frequency band is dynamically analyzed, and multiple target spectra are obtained according to the preset high-frequency fault types. The high-frequency band data can be further divided to obtain more subtle dynamic information. By performing spectral analysis on the high-frequency band signal and combining it with the trend analysis of the time series, the changing trend of the high-frequency signal can be monitored in real time, thereby predicting the following faults in the combustion chamber: incomplete combustion (abnormal fuel-air mixing ratio, uneven fuel distribution), excessive thermoacoustic oscillation (abnormal increase in the amplitude of the thermoacoustic oscillation mode in the combustion chamber), aging of burner components (such as local cracks in the flame tube, blockage or local damage of the combustion nozzle), and abnormal turbulence characteristics (increased turbulence intensity or abnormal turbulence-combustion coupling).

[0082] Specifically, the pressure feature data of each target spectrum in the high-frequency range (such as 40 times the frequency, 50 times the frequency, and up to 200 times the frequency) are extracted, including features such as the maximum amplitude value and the pressure change rate, to obtain a feature set of each spectrum pressure. By performing feature extraction on each target spectrum pressure data, a multi-spectrum pressure feature set is obtained, which can compress a large amount of original pressure data into a representative feature vector, reducing the amount of data while retaining key information related to the fault, and more accurately capturing the detailed information and potential fault characteristics in the pressure signal.

[0083] Specifically, the first pressure prediction model is trained using historical pressure data to obtain a pre-trained first pressure prediction model. Specifically, the first pressure prediction model is an LSTM model. Different spectrum pressure data are predicted according to the pressure feature set to obtain predicted pressure data within spectrum ranges such as 40 times the frequency, 50 times the frequency, and 75 times the frequency. By predicting future pressure data, possible abnormal pressure change trends can be discovered in advance, and an early warning can be issued before the actual abnormality occurs, which helps to take timely measures to prevent the occurrence or expansion of faults.

[0084] Furthermore, for the non-high-frequency band pressure data, the pressure change trend is predicted according to a preset second pressure prediction model to obtain the second pressure change trend, including:

[0085] S501, extracting features based on historical non-high frequency band pressure data to obtain a non-high frequency band pressure feature set;

[0086] S502 : Predicting the pressure using a preset second pressure prediction model according to the non-high-frequency band pressure feature set to obtain a second pressure change trend.

[0087] In this embodiment, the non-high-frequency band pressure data includes a variety of information related to the combustion process. The non-high-frequency band pressure feature set is obtained through feature extraction, and the representative and indicative features of the pressure change trend are mined, including mean, variance, skewness, peak and other features, which can reflect some basic states and laws of the combustion process, such as combustion stability, changes in average pressure level, etc.

[0088] Specifically, the second pressure trend prediction model is an LSTM model. A large amount of historical frequency domain pressure feature data in non-high-frequency bands is used to train the LSTM model to obtain a preset second pressure trend prediction model. The preset second pressure trend prediction model is used to predict the pressure change trend in the non-high-frequency band to obtain the second pressure change trend, which can help operators understand the changes in the non-high-frequency band pressure in the combustion chamber in advance and provide a basis for taking measures in advance.

[0089] Furthermore, the step of issuing an early warning of abnormal pressure conditions based on the multi-band pressure prediction trend includes:

[0090] S601: For a first pressure change trend in a high-frequency band, use a preset abnormal warning model to issue an early warning of pressure abnormality;

[0091] S602: For the second pressure change trend in the non-high frequency band, an early warning of abnormal pressure is issued according to the preset early warning value and high alarm value.

[0092] In this embodiment, different methods are used to detect pressure anomalies in different frequency bands, and targeted anomaly detection can be performed on the pressure characteristics within the frequency band to meet the needs of different frequency band ranges. For high-frequency pressure data, the high-frequency pressure signal will contain some complex dynamic information and potential fault characteristics. This embodiment uses a preset abnormal warning model to identify and classify abnormalities based on machine learning, automatically learn the characteristic pattern of the pressure signal in the high-frequency band, capture the characteristic distribution law of the normal pressure signal in the high-frequency band, and the difference between the abnormal pressure signal and the normal signal, detect the pressure anomaly in the high-frequency band, obtain the pressure anomaly location and anomaly type, and identify the high-frequency pressure anomaly through the abnormal warning model. The high-frequency pressure signal input in real time can be quickly detected to detect abnormalities in a timely manner.

[0093] Specifically, for pressure data in non-high-frequency bands, abnormal pressure conditions are identified by setting abnormal alarm thresholds. Pressure signals in non-high-frequency bands are relatively stable, and their pressure values ​​generally fluctuate within a certain range. By setting pre-alarm and high-alarm values, the pressure signal's value range can be divided into different intervals. When the pressure value exceeds the pre-alarm value, it indicates a potential abnormality that requires attention; when the pressure value exceeds the high-alarm value, it indicates a more serious pressure abnormality and requires immediate action.

[0094] Specifically, the pre-alarm value and the high-alarm value are used to detect abnormal pressure conditions. The pre-alarm value and the high-alarm value can be set based on the understanding of the normal pressure fluctuation range and the anticipation of abnormal conditions. The pre-alarm value and the high-alarm value are set according to the mean and standard deviation of normal pressure fluctuations. The formula is:

[0095]

[0096] Where μ is the mean pressure in the frequency band, P i is the i-th pressure data value in the frequency band, and N is the total number of data points. The standard deviation is calculated as follows:

[0097]

[0098] Where σ is the standard deviation of pressure in the frequency band. The formula for setting the pre-alarm value and high alarm value based on the mean and standard deviation is as follows:

[0099] T 预 =μ+k1×σ;

[0100] T 高 =μ+k2×σ;

[0101] Where, T 预 is the pre-alarm value, T 高 is the high alarm value, and k1 and k2 are set multiples. The values ​​of k1 and k2 can be determined based on actual needs and safety requirements. Generally, the value of k1 should be smaller than the value of k2 to ensure that the pre-alarm value is triggered before the high alarm value. When setting the abnormal threshold, it is necessary to consider the characteristics of the pressure data and the actual application scenario. For example, if the data fluctuates greatly, the values ​​of k1 and k2 need to be appropriately increased; if the application scenario has higher safety requirements, a stricter abnormal threshold needs to be set. The setting of the abnormal threshold is a dynamic process and needs to be adjusted and optimized according to the actual situation. In actual applications, the abnormal threshold can be adjusted by observing the alarm situation, analyzing data changes, etc. to improve the accuracy and reliability of monitoring.

[0102] When the pulsating pressure value exceeds the pre-alarm value but does not reach the high alarm value, the system is considered to be in a warning state and requires close attention. At this time, by locating the warning point in the pulsating pressure change curve, the specific time and frequency of the warning can be determined. By setting the pre-alarm value and high alarm value, a graded warning of abnormal pulsating pressure can be achieved. When the pulsating pressure value is within the warning range, the system can issue a warning signal in advance, reminding the operator to pay attention to the reaction in the combustion chamber and take necessary measures to avoid potential risks.

[0103] When the pulsating pressure value exceeds the high alarm value, the system is considered to be in a serious abnormal state and immediate measures must be taken to avoid failure or accidents. At this time, by locating the alarm point in the pulsating pressure change curve, the specific time and frequency of the alarm can be determined, and an alarm signal can be immediately issued. By setting the high alarm value, the severity of the pulsating pressure anomaly can be determined. When the pulsating pressure value exceeds the high alarm value, the system can immediately issue an alarm signal, reminding the operator to take emergency measures to avoid more serious consequences.

[0104] Further, such as Figure 3 , for the first pressure change trend in the high frequency band, using a preset abnormal warning model to issue an early warning of pressure abnormality, including:

[0105] S701. Draw a pressure change trend curve for each target spectrum pressure according to the first pressure change trend;

[0106] S702: Calculate the slope and amplitude increment of the change trend curve according to the change trend curve to obtain the trend characteristics of each target spectrum pressure;

[0107] S703: Update the alarm threshold using a preset fault spectrum model according to the preset high-frequency fault type and historical fault pressure data to obtain an updated alarm threshold for each spectrum pressure;

[0108] S704: Compare the trend feature with the updated alarm threshold of the corresponding spectrum to obtain the pressure anomaly location and anomaly type;

[0109] S705: Issue an early warning based on the abnormal pressure location and type through a preset abnormal warning model.

[0110] In this embodiment, based on the predicted pressure change trend of each frequency band, a change trend curve of each spectrum pressure is drawn to show the pressure change trend and fluctuation. In the pulsating pressure change curve, different spectrum positions reflect the change of the pressure signal under different frequency components. The slope represents the rate of change of the curve at a certain point, reflecting the speed of change of pressure at that moment. The amplitude increment represents the change in pressure amplitude at adjacent time points or a specific time period. The slope and amplitude increment of the change trend curve at different positions are calculated to obtain the trend characteristics of each spectrum pressure. Different fault types will cause abnormal changes in the slope and amplitude increment of the pressure curve. For example, when a loose fault occurs in the equipment, the slope may increase and the amplitude increment may be abnormal at certain frequency spectrum positions. By calculating the slope and amplitude increment at the preset spectrum position, the abnormal situation can be identified and the abnormal type can be judged.

[0111] Specifically, the alarm thresholds in the preset standards for multiple fault types can be derived based on a large amount of experimental data, actual operating experience, and theoretical analysis. The preset fault spectrum model is trained using historical fault pressure data to obtain a pre-trained fault spectrum model. According to the fault type and fault pressure data, the alarm thresholds in each frequency band are dynamically updated to obtain updated alarm thresholds. The updated alarm thresholds can better adapt to changes in different operating conditions and fault characteristics, reducing false alarms and missed alarms.

[0112] Specifically, based on the updated alarm threshold, the trend characteristics are compared with the corresponding alarm threshold. When the trend characteristics exceed the threshold range, it means that the spectrum pressure is abnormal. According to the correspondence between different high-frequency fault types and spectrum pressure change characteristics, the type of abnormality can be further determined, and the time and position information corresponding to the abnormal spectrum pressure can be recorded to determine the specific location and type of the pressure abnormality. The location and type of the pressure abnormality are promptly communicated to relevant personnel. According to the location and type of the pressure abnormality, through the preset abnormal warning model, the appropriate warning method and content can be selected, so that the operator can quickly locate and handle the abnormality, reduce the fault handling time, and improve the operation efficiency of the combustion system.

[0113] Furthermore, the obtaining of combustion pulsation pressure data in the combustion chamber includes:

[0114] S801, performing windowing processing on the combustion pulsation pressure data to obtain windowed pressure data;

[0115] S802, resampling the windowed pressure data to obtain resampled data;

[0116] S803: Combine the resampled data to obtain updated combustion pulsation pressure data.

[0117] In this embodiment, the combustion pulsation pressure data is windowed to obtain windowed pressure data. Windowing is a method of windowing continuous signals. In time series analysis, data is divided into multiple small blocks by applying a sliding window to the data. Common windows include rectangular windows, triangular windows, Hanning windows, Hamming windows, Gaussian windows, etc., and a suitable window can be selected according to data characteristics and actual needs. In the processing of real-time combustion pulsation pressure data, windowing can help reduce the boundary effect of the data and improve the ability to analyze the local characteristics of the data. Through windowing, the boundaries of the data can be smoothed and the boundary effects caused by data truncation can be reduced. The windowed data can be more focused on the local time period or spatial area, thereby improving the ability to analyze the local characteristics of the combustion pulsation pressure data.

[0118] Specifically, the windowed pressure data is resampled to obtain resampled data. The sampling rate is adaptively adjusted based on the data characteristics to obtain more consistent sampled data. Resampling removes redundant information from the original data, improving data quality and usability. The resampled data is combined to form complete resampled combustion pulsation pressure data, maintaining data integrity and facilitating subsequent analysis of combustion pulsation pressure.

[0119] Furthermore, the resampling process is performed on the windowed pressure data to obtain resampled data, including:

[0120] S901, extracting the windowed pressure data to obtain extracted data;

[0121] S902: Perform interpolation processing on the extracted data to obtain resampled data.

[0122] In this embodiment, the windowed pressure data is decimated using an integer multiple decimation method to obtain decimated data. The integer multiple decimation method is to extract a sampling point every D-1 data from the original sampling sequence x(p), where p is the sampling point order of the original sampling sequence, to form a new sequence xD(m), where:

[0123] xD(m)=x(mD);

[0124] In the formula, D is the decimation factor, and D is a positive integer. This decimation method is equivalent to reducing the data sampling frequency, that is, the spectral period of the signal after decimation is reduced to 1 / D of the original. By adjusting the decimation factor D, the sampling rate of the data can be adjusted to adapt to different application scenarios. Suppose there is an original sampling sequence x(n) = {1, 2, 3, 4, 5, 6, 7, 8}, and the decimation factor D = 2, then the new sequence xD(m) after decimation is {1, 3, 5, 7}; by processing the original sequence using the integer multiple decimation method, data is taken out at equal intervals and reordered, which can significantly reduce the amount of data, while retaining the main features of the signal, reducing the sampling rate to an appropriate processing range, eliminating the interference of out-of-band signal spectra and noise, reducing storage and calculation costs, and facilitating subsequent signal processing and analysis.

[0125] Specifically, the data after decimation is interpolated using the fractional multiple interpolation method to obtain the resampled data. After integer multiple decimation, the sampling rate of the data decreases. To increase the sampling rate of the data, the fractional multiple interpolation method can be used. Through interpolation processing, denser data points can be obtained, thus more accurately describing the changes in the signal. In the fractional multiple interpolation method, the ratio of the resampling frequency to the initial sampling frequency (i.e., the resampling multiple) needs to be converted into the form of a relatively prime integer ratio first, that is:

[0126] L = I / f;

[0127] In the formula, L is the resampling multiple, I is the interpolation multiple, and f is the decimation multiple. First, perform I-fold interpolation on xD(m), and then perform f-fold decimation to obtain the resampled sequence. For example, let the resampling multiple L be 0.75, and the sampling frequency of the initial sampling sequence be 2×10 4 Hz, then the resampling frequency is 1.5×10 4 Hz. During the resampling process, the resampling multiple 0.75 needs to be regularized to 3 / 4 first, and at this time, the interpolation and decimation multiples I and f are obtained. Then, first perform 3-fold interpolation on the initial sampling sequence to obtain a discrete sequence with a sampling frequency of 6×10 4 Hz, and then perform 4-fold decimation to finally obtain a resampled sequence with a sampling frequency of 1.5×10 4 Hz. [[ID=!9]]

[0128] Specifically, the number of sampling points determines the sampling time, and the sampling time is also the insertion position of the sampling value. The number of resampling points is related to the number of resampled values. After the number of resampled values is determined, the sampling time series is also determined. When 0 < L < 1, the change in the number of sampling points before and after resampling is explained in two cases:

[0129] When the last point of the resampled sequence coincides with the last point of the initial sampling sequence, the number of sampling points of its resampled sequence can be calculated by the following formula:

[0130] Nr =L(N0-1)+1;

[0131] Where N0 is the number of initial sampling points; N r is the number of resampling points.

[0132] When the last point of the resampling sequence does not coincide with the last point of the initial sampling sequence, the number of resampling points can be calculated as follows:

[0133] N r =[L(N0-1)]+1;

[0134] Where [L(N0-1)] means rounding down.

[0135] When L>1, the resampling process adopts the method of filling points, and the number of resampling points can be calculated by the following formula:

[0136] N r =[N0×L]+1;

[0137] After the number of resampling points is determined, the resampled data can be expressed as:

[0138]

[0139] Where, t r (n) is the time series data after resampling, n is the sampling point order of the time series data after resampling, f r The resampling frequency is first reduced to a suitable processing range through integer decimation filtering, while eliminating interference from out-of-band signal spectrum and noise. Fractional interpolation filtering is then used to adjust the sampling rate to synchronize with the transmitted symbols, increasing the data sampling rate and making the data smoother and more continuous.

[0140] In other embodiments, real-time correction is performed during the frequency band division process, and the pressure data is divided according to a preset frequency range. The frequency division result obtained is not accurate enough, and data that belongs to the high-frequency pressure range is divided into low-frequency data; the high-frequency pressure data is related to the rapid changes and high-frequency oscillations in the combustion process. These data can be used to capture the fine features of the combustion process. When the high-frequency data is mistakenly divided into the low-frequency band, it will interfere with the judgment of the combustion state; the division process is corrected according to the combustion curve.

[0141] Specifically, the initial high-frequency interval is determined according to the combustion curve, and compared with the divided frequency band interval to determine the interval where the high-frequency band data that was mistakenly classified into the low-frequency band is located. The misclassified data is compared with the combustion curve. According to the overall trend and characteristics of the combustion curve, the numerical value and time position of the misclassified data are preliminarily adjusted to make the data trend match the combustion process.

[0142] The dynamic rate of change of the combustion process is analyzed based on the slope of the combustion curve in adjacent intervals of the mis-segmented data. If the slope of the combustion curve of the mis-segmented data increases, it indicates that the combustion process is changing faster. In this case, the mis-segmented data is compressed to make the data more compact on the time axis. If the slope decreases, indicating that the combustion process is changing slower, the data is stretched to extend the data on the time axis to better reflect the combustion state, and a corrected mis-segmented pressure curve is obtained. The corrected mis-segmented pressure curve replaces the mis-segmented initial pressure curve and is divided into the high-frequency range for subsequent analysis. By performing frequency band correction on the pressure data, the misclassification problem of high-frequency band data can be effectively corrected. The corrected curve helps to more accurately monitor the dynamic changes in the combustion process and promptly detect combustion anomalies.

[0143] Example 2:

[0144] In this example, a cylindrical combustion chamber with a height of 10 meters and a diameter of 5 meters is equipped with pulsation pressure sensors. These sensors collect real-time combustion pulsation pressure data during combustion. A Hanning window is applied to this pulsation pressure data to obtain windowed pressure data. This windowed pressure data is then resampled using a dynamic resampling method to obtain resampled pressure data. The resampled pressure data are then combined to obtain complete resampled combustion pulsation pressure data.

[0145] According to the resampled combustion pulsating pressure data, the frequency domain data is segmented, and the frequency range of 300-2500Hz is set as the non-high frequency band, and the frequency range of 2500-5000Hz is set as the high frequency band. The pre-alarm value and high alarm value are set in each frequency band as shown in the following table:

[0146] frequency band Frequency / Hz Pre-alarm value / kPa High alarm value / kPa Non-high frequency band 300～2500 17.24 24.13 High frequency band 2500～5000 9.07 12.45

[0147] A pressure anomaly detection method is used to detect pressure anomalies in different frequency bands in the combustion chamber. In each frequency band, a preset pressure trend prediction model is used to predict the pressure trend, and a pressure change curve is drawn. When the pressure pulsation value is lower than the pre-alarm value, no alarm is given; when the pressure pulsation value is higher than the pre-alarm value and lower than the high alarm value, an early warning is issued to remind staff to pay attention to the reaction situation in the combustion chamber; when the pressure pulsation value is higher than the high alarm value, an alarm is issued, and staff take emergency measures, such as direct shutdown.

[0148] The abnormal alarm threshold of the abnormal type is dynamically updated based on the historical data of each frequency doubling pressure in the high-frequency band to obtain the updated abnormal threshold. For different specific frequency doubling positions in the high-frequency band, the slope or amplitude increment of the pressure change trend is calculated and compared with the updated abnormal threshold related to different faults. When the trend indicator exceeds the specific threshold, the alarm mechanism is triggered. The fault type and judgment criteria are as follows:

[0149] Incomplete combustion: The high-frequency energy continues to increase within a specific frequency range (e.g., 40x to 60x), the slope exceeds the set threshold (e.g., >5% / min), and the broadband random noise energy increases significantly.

[0150] Excessive thermoacoustic oscillation: The amplitude of a specific frequency spectrum in the high-frequency band (such as the 50th and 75th frequencies) increases rapidly, reaching the critical value of the combustion chamber thermoacoustic mode, and the trend slope exceeds 10% / min;

[0151] Local cracks in the flame tube: New frequency components or a sharp increase in amplitude appear in the high-frequency spectrum (such as 80 times to 100 times the frequency), and the spectrum characteristics show periodic fluctuations;

[0152] Abnormal turbulence characteristics: the randomness of the overall energy distribution in the high-frequency band increases, the spectral energy concentration decreases (the specific frequency peak disappears or the distribution expands), or the trend slope remains positive.

[0153] Based on the pressure change trend, the location and type of pressure anomalies can be judged, and operators can be promptly reminded of abnormal conditions in the combustion chamber. This can avoid potential safety risks, improve the supervision of the combustion system, and thus reduce the chance of accidents.

[0154] Example 3:

[0155] Combustion pulsation pressure warning system, such as Figure 4 , used to implement the combustion pulsation pressure trend monitoring method, including:

[0156] A data acquisition module, which acquires combustion pulsation pressure data in the combustion chamber;

[0157] A pressure change trend prediction module predicts the pressure change trends of different frequency bands in the combustion chamber based on the combustion pulsation pressure data to obtain a multi-band pressure prediction trend;

[0158] The early warning module issues an early warning of abnormal pressure conditions based on the multi-band pressure prediction trend.

[0159] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A combustion pulsation pressure trend monitoring method, characterized in that: include: Obtain combustion pulsation pressure data in the combustion chamber; According to the combustion pulsation pressure data, the pressure variation trends of different frequency bands in the combustion chamber are predicted respectively to obtain a multi-band pressure prediction trend; Based on the multi-band pressure prediction trend, an early warning is issued for abnormal pressure conditions; The method of predicting the pressure change trends of different frequency bands in the combustion chamber according to the combustion pulsation pressure data to obtain a multi-band pressure prediction trend includes: Perform frequency domain conversion on the combustion pulsation pressure data to obtain frequency domain pressure data; Dividing the frequency domain pressure data into a plurality of frequency bands according to a preset frequency band range, wherein the plurality of frequency bands include a high frequency band and a non-high frequency band; For high-frequency pressure data, a preset first pressure prediction model is used to predict the pressure change trend to obtain a first pressure change trend; For non-high-frequency band pressure data, a pressure change trend prediction is performed according to a preset second pressure prediction model to obtain a second pressure change trend; The method of predicting the pressure change trend of the high-frequency pressure data using a preset first pressure prediction model to obtain the first pressure change trend includes: According to the preset high-frequency fault types, multiple target frequency spectra are obtained, wherein the high-frequency fault types include incomplete combustion, excessive thermoacoustic oscillation, local cracks in the flame tube, and abnormal turbulence characteristics; dividing the high-frequency band pressure data according to the multiple target spectra to obtain multi-spectrum pressure data; Based on the historical high-frequency pressure data, feature extraction is performed on each spectrum pressure data to obtain a multi-spectrum pressure feature set; According to the multi-spectral pressure feature set, each spectral pressure is predicted respectively using a preset first pressure prediction model to obtain a first pressure change trend; For the non-high-frequency band pressure data, performing pressure change trend prediction according to a preset second pressure prediction model to obtain a second pressure change trend includes: Feature extraction is performed based on historical non-high frequency band pressure data to obtain a non-high frequency band pressure feature set; Predicting the pressure using a preset second pressure prediction model based on the non-high-frequency band pressure feature set to obtain a second pressure change trend; The step of issuing an early warning of abnormal pressure conditions based on the multi-band pressure prediction trend includes: For the first pressure change trend in the high-frequency band, a preset abnormal warning model is used to issue an early warning of pressure anomalies; For the second pressure change trend in the non-high frequency band, an early warning of abnormal pressure conditions is issued based on the preset early warning value and high alarm value.

2. The combustion pulsation pressure trend monitoring method according to claim 1, characterized in that: The method of using a preset abnormal warning model to issue an early warning of pressure abnormality for the first pressure change trend in the high frequency band includes: According to the first pressure change trend, a change trend curve of each target spectrum pressure is drawn; According to the change trend curve, the slope and amplitude increment of the curve are calculated to obtain the trend characteristics of each target spectrum pressure; According to the preset high-frequency fault type and historical fault pressure data, the alarm threshold is updated using the preset fault spectrum model to obtain the updated alarm threshold for each spectrum pressure; Comparing the trend feature with the updated alarm threshold of the corresponding spectrum to obtain the pressure anomaly location and anomaly type; According to the abnormal pressure location and abnormal type, an early warning is issued through a preset abnormal warning model.

3. The combustion pulsation pressure trend monitoring method according to claim 1, characterized in that: The obtaining of combustion pulsation pressure data in the combustion chamber includes: Performing windowing processing on the combustion pulsation pressure data to obtain windowed pressure data; resampling the windowed pressure data to obtain resampled data; The resampled data are combined to obtain updated combustion pulsation pressure data.

4. The combustion pulsation pressure trend monitoring method according to claim 3, characterized in that: The resampling process is performed on the windowed pressure data to obtain resampled data, comprising: Extracting the windowed pressure data to obtain extracted data; Interpolation processing is performed on the extracted data to obtain resampled data.

5. Combustion pulsation pressure warning system, characterized by: A method for monitoring combustion pulsation pressure trends according to any one of claims 1 to 4, comprising: A data acquisition module, which acquires combustion pulsation pressure data in the combustion chamber; A pressure change trend prediction module predicts the pressure change trends of different frequency bands in the combustion chamber based on the combustion pulsation pressure data to obtain a multi-band pressure prediction trend; The early warning module issues an early warning of abnormal pressure conditions based on the multi-band pressure prediction trend.

Images