Intelligent diagnosis method for deep peak regulation fault of circulating fluidized bed boiler

Abstract

The invention discloses an intelligent diagnosis method for a deep peak regulation fault of a circulating fluidized bed boiler, and relates to the field of fault diagnosis, and the method comprises the steps: obtaining key fault sensitive features of the circulating fluidized bed boiler when the circulating fluidized bed boiler is in a deep peak regulation working condition; according to the key fault sensitive features, a multi-fault coupling analysis model is constructed, and fault probability distribution is obtained through the multi-fault coupling analysis model; establishing a boiler dynamic characteristic prediction model according to the long short-term memory network, and inputting the fault probability distribution and the space-time coupling characteristics into the boiler dynamic characteristic prediction model to obtain a three-dimensional diagnosis result; according to the three-dimensional diagnosis result and an adaptive threshold algorithm, generating a graded early warning signal; and starting a local fault tracing module based on transfer learning according to the graded early warning signal to realize fault positioning. The accuracy of fault diagnosis can be improved.

Description

Technical Field

[0001] The present application relates to the field of fault diagnosis, and in particular to an intelligent diagnosis method for deep peak regulation faults of circulating fluidized bed boilers. Background Art

[0002] Circulating fluidized bed boilers (CFBs) are highly efficient and environmentally friendly combustion equipment widely used in power generation, heating, and industrial production. They offer advantages such as strong fuel adaptability, high combustion efficiency, and low pollutant emissions. Deep peak shaving refers to adjusting boiler operating parameters to accommodate rapid fluctuations in grid load when thermal power units are operating at low load. Under deep peak shaving conditions, problems can arise, such as fluidization stability within the furnace, nitrogen oxide emission control, and localized overheating within the furnace. These problems can lead to deteriorating combustion, equipment damage, or even shutdown, seriously impacting boiler operating efficiency and safety.

[0003] Currently, fault diagnosis for deep peak-shaving fluidized bed boilers relies primarily on empirical judgment and simple threshold determination. However, existing methods have low fault diagnosis accuracy and are difficult to adapt to the complex operating environment of deep peak-shaving conditions. Summary of the Invention

[0004] The purpose of this application is to provide an intelligent diagnosis method for deep peak-shaving faults of circulating fluidized bed boilers, which can improve the accuracy of deep peak-shaving fault diagnosis of fluidized bed boilers.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides an intelligent diagnosis method for deep peak-shaving faults in a circulating fluidized bed boiler, comprising:

[0007] When the circulating fluidized bed boiler is in deep peak load regulation, key fault sensitive features of the circulating fluidized bed boiler are obtained;

[0008] Constructing a multi-fault coupling analysis model based on the key fault sensitive characteristics, and obtaining a fault probability distribution through the multi-fault coupling analysis model;

[0009] Establishing a boiler dynamic characteristics prediction model based on a long short-term memory network, inputting the fault probability distribution and spatiotemporal coupling characteristics into the boiler dynamic characteristics prediction model to obtain a three-dimensional diagnosis result;

[0010] generating a graded warning signal based on the three-dimensional diagnostic results and an adaptive threshold algorithm;

[0011] According to the graded warning signal, a local fault tracing module based on transfer learning is started to realize fault location.

[0012] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0013] The present application provides an intelligent diagnosis method for deep peak-shaving faults of circulating fluidized bed boilers, which obtains the key fault-sensitive characteristics of the circulating fluidized bed boiler when the circulating fluidized bed boiler is in a deep peak-shaving condition. Based on the key fault-sensitive characteristics, a multi-fault coupling analysis model is constructed, and the fault probability distribution is obtained through the multi-fault coupling analysis model. A boiler dynamic characteristic prediction model is established based on a long short-term memory network, and the fault probability distribution and the spatiotemporal coupling characteristics are input into the boiler dynamic characteristic prediction model to obtain a three-dimensional diagnosis result; based on the three-dimensional diagnosis result and the adaptive threshold algorithm, a graded warning signal is generated; based on the graded warning signal, a local fault tracing module based on transfer learning is started to achieve fault location. The present application generates a graded warning signal based on the three-dimensional diagnosis result and the adaptive threshold algorithm, and further based on the graded warning signal, a local fault tracing module based on transfer learning is started to locate the fault location, thereby improving the accuracy of the deep peak-shaving fault diagnosis of the fluidized bed boiler. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 This is a diagram of the application environment of an intelligent diagnosis method for deep peak-shaving faults of a circulating fluidized bed boiler in one embodiment of the present application;

[0016] Figure 2 A flow chart of an intelligent diagnosis method for deep peak-shaving faults in a circulating fluidized bed boiler provided in one embodiment of the present application;

[0017] Figure 3 A schematic diagram of the functional modules of an intelligent diagnostic device for deep peak-shaving faults in a circulating fluidized bed boiler provided in one embodiment of the present application;

[0018] Figure 4 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0020] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0021] The intelligent diagnosis method for deep peak regulation fault of circulating fluidized bed boiler provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store data that server 104 needs to process. The data storage system can be set up separately, integrated with server 104, or placed in the cloud or on other servers. Terminal 102 can send the key fault-sensitive features to be processed to server 104. After receiving the key fault-sensitive features to be processed, server 104 constructs a multi-fault coupling analysis model based on the key fault-sensitive features of the circulating fluidized bed boiler, and obtains a fault probability distribution through the multi-fault coupling analysis model. A boiler dynamic characteristics prediction model is established based on a long-short-term memory network. The fault probability distribution and spatiotemporal coupling features are input into the boiler dynamic characteristics prediction model to obtain a three-dimensional diagnosis result. Based on the three-dimensional diagnosis result and an adaptive threshold algorithm, a graded warning signal is generated. Based on the graded warning signal, a local fault tracing module based on transfer learning is activated to achieve fault location. Server 104 can feedback the obtained fault location to terminal 102. In addition, in some embodiments, the intelligent diagnosis method for deep peak-shaving faults of circulating fluidized bed boilers can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly process the key fault sensitive features to be processed, or the server 104 can obtain the key fault sensitive features to be processed from the data storage system and process the key fault sensitive features to be processed.

[0022] Terminal 102 may include, but is not limited to, various desktop computers, laptops, smartphones, tablet computers, IoT devices, and portable wearable devices. IoT devices may include smart speakers, smart TVs, smart air conditioners, and smart car devices. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. Server 104 may be implemented as a standalone server or a server cluster consisting of multiple servers, or may be a cloud server.

[0023] Deep peak shaving refers to a method in which generator sets reduce their power generation to a level significantly below their rated power, even approaching or reaching their minimum stable combustion load, to accommodate large fluctuations in grid load, particularly during periods of low grid load. Deep peak shaving requires generator sets to reduce their power to 30% or even less of their rated power. For example, a 300MW rated unit may need to be reduced to 90MW or even less during deep peak shaving. Under deep peak shaving conditions, load fluctuations are both large and frequent, requiring the generator set to respond quickly to these load changes. At night, especially during periods of low residential electricity demand, grid load demand drops significantly. During these times, generator sets must perform deep peak shaving to avoid excess power and grid instability. During holidays and other periods of low industrial production, load demand decreases significantly, requiring generator sets to adapt to these fluctuations. With the rapid development of renewable energy sources such as wind and solar, their generation has become intermittent and unpredictable. To ensure stable grid operation, thermal power plants must perform deep peak shaving to fill the gaps in renewable energy generation.

[0024] Deep peak shaving of circulating fluidized bed boilers refers to reducing the power generation capacity to a level far below its rated power by adjusting the operating parameters and control strategies of the boilers during periods of low load demand on the power system to meet the peak shaving needs of the power grid. This mode of operation requires the boiler to be able to quickly reduce the load in a short period of time and quickly restore the load when necessary to ensure the stable operation of the power grid. Circulating fluidized bed boilers face many problems when operating in deep peak shaving mode. For example, the boiler needs to maintain a stable combustion state at low load while controlling pollutant emissions, especially nitrogen oxides (NO x In addition, deep peak load regulation also requires the boiler to have good hydrodynamic safety to prevent hydrodynamic instability during low-load operation.

[0025] In an exemplary embodiment, Figure 2 As shown, an intelligent diagnosis method for deep peak regulation faults of circulating fluidized bed boilers is provided. The method is executed by a computer device, specifically a computer device such as a terminal or a server, or a terminal and a server. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate the process, including the following steps S201 to S205.

[0026] In step S201, when the circulating fluidized bed boiler is in a deep peak-shaving operating condition, key fault-sensitive features of the circulating fluidized bed boiler are obtained.

[0027] In one embodiment, the step S201 of "obtaining key fault-sensitive features of the circulating fluidized bed boiler" includes the following sub-steps S2011-S2014:

[0028] S2011, using multimodal sensors to collect the initial operating parameters of the circulating fluidized bed boiler in real time, including the boiler load change rate ΔP, the bed fluidization velocity θ fluid and the separator pressure difference ΔP sep .

[0029] S2012: Normalize the initial operating parameters.

[0030] The normalization formula is:

[0031]

[0032] Among them, x represents the initial operating parameters, x max and x min Represent the maximum and minimum values of the initial operating parameters, respectively. The collected initial operating parameters are normalized to ensure that different types of parameters can be compared on the same scale. The data are mapped to the [0, 1] interval to eliminate differences in the dimensions and numerical ranges of different parameters.

[0033] Normalized boiler composite change rate ΔP norm Calculated by the following formula:

[0034]

[0035] Normalized bed fluidization velocity θ fluid,norm Calculated by the following formula:

[0036]

[0037] Normalized separator pressure difference ΔP sep,norm Calculated by the following formula:

[0038]

[0039] S2013. Extracting time domain features and frequency domain features from the normalized initial operating parameters, wherein time domain feature extraction refers to calculating the fluctuation standard deviation and mutation slope of the normalized initial operating parameters, and frequency domain feature extraction refers to calculating the spectral energy distribution of the acoustic emission signal of the normalized initial operating parameters.

[0040] Specifically, the standard deviation of fluctuation is a measure of the degree of fluctuation of a parameter over time. The standard deviation of fluctuation is expressed by the following formula:

[0041]

[0042] Among them, x i represents the normalized initial operating parameters, represents the average value of the normalized initial running parameters, and N represents the number of data points.

[0043] The sudden change of parameter value can be obtained by mutation slope, which can be calculated by the slope between adjacent time points. The mutation slope is expressed by the following formula:

[0044]

[0045] Wherein, Δx represents the change of the normalized initial operating parameters, and Δt represents the time change.

[0046] The normalized acoustic emission signal is subjected to frequency domain feature extraction, and the spectrum energy distribution of the acoustic emission signal is calculated by frequency domain feature extraction. The spectrum energy distribution can be obtained by short-time Fourier transform. For example, the formula for calculating the spectrum energy distribution using short-time Fourier transform is as follows:

[0047] F freq =|STFT(x)| 2 ;

[0048] Wherein, STFT(x) is the result of short-time Fourier transform of the normalized initial operating parameter x.

[0049] S2014. Dynamically weight the time domain features and the frequency domain features to obtain operating parameters of the circulating fluidized bed boiler.

[0050] In one embodiment, step S2014 includes the following sub-steps S20141-S20142:

[0051] S20141, according to the normalized boiler composite change rate ΔP norm , normalized bed fluidization velocity θ fluid,norm And the normalized separator pressure difference ΔP sep,norm Calculate dynamic weighting coefficients.

[0052] The calculation formula of the dynamic weighting coefficient is as follows:

[0053]

[0054] Where a represents the preset weight coefficient of boiler load change rate, b represents the preset weight coefficient of bed fluidization velocity, and c represents the preset weight coefficient of separator pressure difference; ΔP norm represents the normalized composite rate of change of the boiler, θ fluid,norm represents the normalized bed fluidization velocity, ΔP sep,normrepresents the normalized separator pressure difference.

[0055] S20142. According to the dynamic weighting coefficient, the dynamically weighted fluctuation standard deviation, the dynamically weighted mutation slope and the dynamically weighted frequency domain characteristics are calculated. The dynamically weighted fluctuation standard deviation, the dynamically weighted mutation slope and the dynamically weighted frequency domain characteristics are calculated as the key fault sensitive characteristics of the circulating fluidized bed boiler.

[0056] Specifically, the dynamically weighted volatility standard deviation is calculated using the following formula:

[0057] F std,weighted =ω i ·F std ;

[0058] The dynamic weighted mutation slope is calculated by the following formula:

[0059] F slope,weighted =ω i ·F slope ;

[0060] The frequency domain features after dynamic weighting are calculated using the following formula:

[0061] F freq,weighted =ω i ·F freq .

[0062] In step S202 , a multi-fault coupling analysis model is constructed based on key fault sensitivity characteristics, and a fault probability distribution is obtained through the multi-fault coupling analysis model.

[0063] In one embodiment, the step S202 of "building a multi-fault coupling analysis model based on key fault sensitivity characteristics" includes the following sub-steps S2021-S2023:

[0064] S2021. Divide the operating state of the boiler into multiple cluster subspaces based on a fuzzy clustering algorithm, each cluster subspace representing a typical operating mode or a fault precursor state.

[0065] S2022. For each cluster subspace, perform similarity matching between the key fault sensitive features and the fault feature library, and output a preliminary fault probability distribution based on the matching results.

[0066] Specifically, the weights of the time domain and frequency domain features are dynamically adjusted according to the current operating status of the boiler (such as load change rate, fluidization wind speed). The weights are calculated by fuzzy logic or adaptive algorithm, for example: 时域 =f(load change rate, fluidizing wind speed). The weighted combination is expressed by the following formula:

[0067] F 加权 =w时域 ·F 时域 +w 频域 ·F 频域 ;

[0068] Among them, F 时域 and F 频域 represent the time domain eigenvector and frequency domain eigenvector respectively.

[0069] The preliminary fault probability refers to the probability output after fusing multiple pieces of evidence through the DS synthesis rule and the Dempster combination rule. The probability output is expressed as: P = {coking: 0.55, wear: 0.3, hood blockage: 0.1, unknown: 0.05}. This distribution reflects the current confidence level of each fault hypothesis and is used for subsequent corrections.

[0070] The preprocessed feature parameters are then matched against a pre-set fault feature library for similarity. This library, established through analysis and summarization of a large amount of historical fault data, contains feature vectors corresponding to different fault types. The DS evidence theory is used to calculate the basic probability distribution function for each fault type, outputting a preliminary fault probability distribution. The boiler operating status is shown in Table 1:

[0071] Table 1

[0072]

[0073] The improved fuzzy clustering algorithm dynamically divides the operating state into multiple subspaces, each of which represents a typical operating mode or fault precursor state. The subspace division is based on a combination of time domain features, frequency domain features, and operating parameters. Time domain features represent the standard deviation of fluctuations (reflecting the severity of parameter fluctuations) and the mutation slope (capturing parameter mutation events). Frequency domain features represent the spectral energy distribution of the acoustic emission signal (identifying the vibration mode of a specific fault). Operating parameters represent the boiler load change rate, bed fluidization velocity, and separator pressure difference. Each subspace corresponds to an operating mode or fault type of the boiler. Through dynamic division, the model can: independently analyze fault characteristics in different subspaces to avoid the complexity of the global model. Combined with DS evidence theory, more accurate fault probability calculations are performed for the features within the subspace.

[0074] The fault feature library is a pre-built mapping table of fault types and typical feature vectors, which is used to match real-time features. The feature vector of each fault type is determined by historical data or simulation analysis. When constructing a multi-fault coupling analysis model, similarity matching is performed based on the DS evidence theory. First, a preset fault feature library must be established. This library is obtained by analyzing and summarizing a large amount of historical fault data, and contains feature vectors corresponding to different fault types. Then, the feature parameters after dynamic weighted preprocessing are compared with the feature vectors in the fault feature library. The DS evidence theory is used to calculate the similarity between each sample feature parameter and the feature vector of each fault type. This calculation process will derive the basic probability distribution function of each fault type, thereby determining the possibility that the sample belongs to each fault type and completing the similarity matching.

[0075] The output of the similarity matching calculation is a preliminary fault probability distribution. This distribution presents the probability of various preset fault types occurring under the current boiler operating state in the form of probabilities. For example, if at a certain moment, the calculated probability of a "material blockage" fault in the boiler is 0.6, the probability of a "heating surface wear" fault is 0.3, and the total probability of all other faults is 0.1, then this set of probability values constitutes the preliminary fault probability distribution for the boiler's operating state at that moment. This preliminary fault probability distribution provides initial data for further correction of fault probabilities using the adaptive weighted fusion algorithm, helping to more accurately determine whether a boiler fault exists and what type of fault may occur.

[0076] S2023. Dynamically modify the preliminary fault probability distribution through an adaptive weighted fusion algorithm. Multiple clustering subspaces, the preliminary fault probability distribution, and the adaptive weighted fusion algorithm constitute the multi-fault coupling analysis model.

[0077] Specifically, the adaptive weighted fusion algorithm aims to more accurately fusion-correct the initial fault probability by dynamically adjusting the weight coefficients, thereby improving the accuracy of fault diagnosis. In the diagnosis of deep peak-shaving faults in circulating fluidized bed boilers, the algorithm adjusts the weights based on the correlation between the fault evolution patterns in the boiler's historical peak-shaving data and the current load rate.

[0078] The weight coefficient of the adaptive weighted fusion algorithm is dynamically adjusted through a linear regression model. The regression model is based on the fault evolution law in the historical peak-shaving data of the boiler and the current load rate r load The correlation between , and its regression equation is:

[0079] w j =β0+β1r load +β2Fault_Severity j ;

[0080] Among them, wj is the weight of the jth fault type, β0, β1, β2 are regression coefficients, Fault_Severity j As the historical fault severity index, the preliminary failure rate is fused according to the adjusted weight coefficient to obtain a more accurate failure probability distribution.

[0081] A large amount of historical peak-shaving data of boilers is collected, covering information such as operating parameters, fault occurrence and corresponding fault severity under different working conditions. At the same time, the current load rate data is obtained in real time. The linear regression model is trained using historical peak-shaving data, and the regression coefficients β0, β1, and β2 are solved by methods such as least squares. During the training process, the current load rate and historical fault severity indicators are used as independent variables, and the fault type weight is used as the dependent variable, so that the model can learn the relationship between them. According to the trained model, the weight coefficient of each fault type is calculated in combination with the current load rate and historical fault severity indicators obtained in real time. Then, these weight coefficients are used to perform weighted fusion on the preliminary fault probability obtained based on DS evidence theory. Assume that the probability of the j-th fault type in the preliminary fault probability distribution is P j , the corrected failure probability P ' J The calculation method is P ' J =w j ×P j . In this way, the dynamic correction of the preliminary failure probability is achieved, so that the failure probability more accurately reflects the actual situation. The system obtains the current load rate data in real time and combines it with the stored historical peak-shaving data of the boiler. During operation, new current load rate data and corresponding fault conditions (such as whether a fault has occurred, the type of fault, the severity, etc.) are continuously included in the analysis. Using these new data, the linear regression model is retrained, and the regression coefficients β0, β1, and β2 are updated to achieve dynamic adjustment of the weight coefficient. For example, when the boiler load changes frequently and the amplitude is large, the new load rate data will prompt the model to recalculate the weight coefficient, so that fault diagnosis can adapt to changes in operating conditions more promptly.

[0082] The online learning content of the boiler's historical peak-shaving data includes the operating parameters of the boiler in different peak-shaving stages. The operating parameters include but are not limited to bed temperature distribution, separator pressure difference, fluidization wind speed, and flue gas component concentration. It also contains detailed information on each fault, such as the time of fault occurrence, fault type, fault handling process and results, as well as fault-related operating condition information, such as the load level at the time and load change trend.

[0083] The content of fault evolution law covers the signs before the fault occurs, such as the trend of certain operating parameters gradually deviating from the normal range. During the development of the fault, the change pattern of each parameter, for example, how other related parameters respond when the bed temperature changes suddenly, whether the severity of the fault gradually increases or remains stable over time, etc., as well as the mutual influence and transformation relationship between different fault types, for example, whether material blockage will cause failures in other components. The content of current load rate refers to the speed of change of the current load of the boiler, which is usually measured by the change in load per unit time, such as megawatts / minute. It reflects the speed of boiler load adjustment. Under deep peak-shaving conditions, the load rate changes frequently and with large amplitudes, which is an important factor affecting the accuracy and timeliness of fault diagnosis.

[0084] In one embodiment, the step S2021 of "building a multi-fault coupling analysis model based on key fault sensitivity characteristics" includes the following sub-steps S20211-S20213:

[0085] S20211. Initialize cluster centers and fuzzy coefficients.

[0086] The value of the fuzzy coefficient m is usually between 1.5 and 2.5.

[0087] S20212. Based on the initialized cluster centers and fuzzy coefficients, iteratively calculate the membership of the samples to each initialized cluster center.

[0088] Each sample x i To each cluster center v j The membership degree u ij Calculated by the following formula:

[0089]

[0090] in, Update the cluster center according to the membership degree until the cluster center converges; x i represents the i-th sample, v j represents the jth cluster center, c represents the number of clusters, and m represents a constant greater than 1, called the fuzzification parameter, which is used to control the fuzziness of the membership. The larger m is, the more fuzzy the distribution of the membership; the closer m is to 1, the clearer the distribution of the membership.

[0091] S20213. Update the initialized cluster centers according to the membership degree until the cluster centers converge.

[0092] Repeat the above steps S20211-S20213 until the cluster center converges, that is, the change between the two cluster centers is less than the set threshold.

[0093] The corresponding relationship between the fault type and characteristic vector of the circulating fluidized bed boiler is shown in Table 2:

[0094] Table 2

[0095]

[0096] In step S203, a boiler dynamic characteristic prediction model is established based on the long short-term memory network, and the fault probability distribution and the spatiotemporal coupling characteristics are input into the boiler dynamic characteristic prediction model to obtain a three-dimensional diagnosis result.

[0097] In one embodiment, step S203 includes the following sub-steps S2031-S2034:

[0098] S2031. Calculate the standard deviation of oxygen fluctuations through a sliding window to obtain a combustion stability index in a time dimension, where the combustion stability index is used to reflect the stability of the combustion state of the boiler during continuous operation.

[0099] Within a selected time window (e.g., 5 minutes), the oxygen content data of the boiler is collected in real time. The standard deviation of the oxygen content fluctuation within this time window is calculated using the following formula:

[0100]

[0101] Among them, x i Indicates the oxygen measurement value at each time point, represents the average of these measurements, and N represents the number of samples. The lower the standard deviation, the more stable the combustion process. This index reflects the stability of the combustion process; the larger the index, the worse the combustion stability.

[0102] S2032. Perform spatial correlation analysis on the data collected by the bed pressure sensor array to obtain a bed fluidization uniformity index in the spatial dimension. The bed fluidization uniformity index is used to evaluate whether there are local anomalies in the coordination of bed material fluidization in different regions.

[0103] Specifically, multiple bed pressure sensors are arranged in an array on the boiler bed to collect bed pressure data at each location in real time. For every two sensors in the sensor array, bed pressure data within the same time period are selected, and the correlation between the bed pressure data of each two sensors within the same time period is calculated using the Pearson correlation coefficient formula:

[0104]

[0105] Among them, X i and Y i Represents the readings of two sensors at the same time, and Respectively represent the average value of each reading.

[0106] The above calculations are performed for all pairwise combinations of sensors, resulting in a series of correlation coefficient values. Correlation coefficients range from -1 to 1. Values close to 1 indicate highly consistent bed pressure trends at the two sensor locations, indicating relatively uniform fluidization. Values close to 0 indicate little linear relationship and the presence of potentially uneven fluidization. Values close to -1 indicate opposite pressure trends at the two locations, which is uncommon under normal circumstances.

[0107] The correlation coefficient values between all sensors are combined to evaluate the fluidization uniformity of the entire bed. If the correlation coefficients between most sensors are above 0.8, the overall fluidization uniformity of the bed is good. If many correlation coefficients are below 0.5, it indicates that the fluidization may be uneven, which requires further analysis and treatment.

[0108] S2033. Construct the boiler dynamic characteristics prediction model based on long short-term memory network and attention mechanism.

[0109] S2034. Input the spatiotemporal coupling characteristics and the fault probability distribution into the boiler dynamic characteristic prediction model, and output a three-dimensional diagnosis result. The three-dimensional diagnosis result includes the fault type, severity, and development trend.

[0110] Specifically, a long short-term memory (LSTM) network with at least two hidden layers is constructed, with each layer containing 128 memory units. An attention mechanism is introduced to weight the input features to highlight key abnormal features. The attention weight is calculated using the following formula:

[0111] α t =softmax(W h h t +W s S t );

[0112] Among them, W h and W s represents the trainable parameter matrix, h t represents the state vector of the LSTM hidden layer, S t represents the context vector.

[0113] The LSTM model is trained using historical operating data, and model parameters are optimized to accurately learn the relationship between fault characteristics and fault type, severity, and evolution trends. The corrected fault probability distribution and spatiotemporal coupling characteristics (such as the combustion stability index and bed fluidization uniformity index) are fed into the trained LSTM model as input. The model outputs three-dimensional diagnostic results, including fault type (identifying the specific fault category), severity (expressing the degree of harm caused by the fault as a numerical value or level), and evolution trend (predicting the future direction of the fault). This information helps operators promptly identify potential problems and take appropriate measures to prevent them from escalating, thereby improving boiler safety and economic efficiency.

[0114] In step S204, a graded warning signal is generated based on the three-dimensional diagnosis result and the adaptive threshold algorithm.

[0115] The adaptive threshold algorithm dynamically adjusts the threshold based on the statistical distribution of historical fault data using the following formula:

[0116]

[0117] Among them, μ history and σ history They represent the mean and standard deviation of historical fault data respectively, k is the preset confidence coefficient, and the value range is 0.5≤k≤2, N history and N current They are the number of historical samples and the number of current samples. When the diagnosis result exceeds the corresponding threshold, different levels of warnings are triggered.

[0118] In this disclosure, the early warning signals are divided into three levels according to the severity of the fault and its impact on the operation of the boiler: the first-level early warning is for minor faults, such as small fluctuations in parameters close to the abnormal threshold, which have little impact on operation but still require attention. At this time, a yellow warning mark is displayed on the monitoring interface and a prompt sound is used to remind you; the second-level early warning corresponds to moderate faults, and some operating parameters have exceeded the normal range, which may cause a decrease in combustion efficiency or an increase in energy consumption. The early warning signal is enhanced, and an orange mark is displayed accompanied by a rapid prompt sound. At the same time, relevant personnel are notified by text message or email to check and deal with it in time; the third-level early warning is the highest level, which is used to deal with faults that seriously threaten the safe operation of the boiler, such as sudden changes in bed temperature, serious abnormal fluidization, etc. The system will display a red warning mark, issue a high-decibel alarm, and automatically cut off the operation of some non-critical equipment to prevent the fault from expanding. At the same time, all relevant personnel are urgently notified to intervene quickly.

[0119] Current operating parameters refer to various key parameters collected in real time during the operation of the boiler, mainly including operating status parameters (such as bed temperature distribution, separator pressure difference, fluidization air velocity, flue gas component concentration, etc.). These parameters can reflect the real-time operating status of the boiler, and their abnormal fluctuations are an important basis for fault diagnosis. For example, a sudden rise or fall in bed temperature may indicate abnormal combustion or fluidization state instability; they also include load-related parameters, such as the boiler load change rate and the current load value. Under deep peak-shaving conditions, frequent and drastic load changes will affect the stability of equipment operation. Excessive load change rates may induce multiple faults. Therefore, they are important reference indicators for early warning and diagnosis; in addition, they also include equipment operating parameters, such as coal feed rate and air volume. These parameters directly affect the combustion efficiency and energy conversion process of the boiler. Their abnormal adjustments may lead to problems such as incomplete combustion and increased wear of the heating surface. When performing fault identification and adaptive threshold adjustment, the above-mentioned parameters need to be comprehensively analyzed and judged.

[0120] In step S205 , according to the graded warning signal, a local fault tracing module based on transfer learning is started to locate the fault.

[0121] In one embodiment, step S205 includes the following sub-steps S2051-S2056:

[0122] S2051. Determine whether there is a sudden change in bed temperature or abnormal fluidization based on the graded warning signal.

[0123] Specifically, the system first monitors the boiler's operating status in real time. A key component of this monitoring is the generation of graded early warning signals. These signals are generated based on changes in various parameters during boiler operation, providing early warning of potential anomalies. By analyzing these graded early warning signals, the system can determine whether there are two common precursors to localized failures: sudden changes in bed temperature or fluidization anomalies. A sudden change in bed temperature may indicate an unstable combustion process, while an abnormal fluidization may indicate a problem with material circulation or distribution, both of which could threaten the safe operation of the boiler.

[0124] S2052: If there is a sudden change in bed temperature or the fluidization is abnormal, activate the local fault tracing module based on transfer learning.

[0125] Specifically, once a sudden change in bed temperature or abnormal fluidization is detected, the system takes immediate action. At this point, the system activates a local fault tracing module based on transfer learning. Transfer learning is an advanced machine learning technique that can quickly adapt and apply to new tasks or scenarios based on existing domain knowledge. In this scenario, the local fault tracing module based on transfer learning can leverage the experience and knowledge accumulated in previous similar boiler operating environments to more efficiently trace the source of the current local fault, providing strong support for subsequent fault handling.

[0126] S2053. Collect the acoustic emission signal at the current moment through the acoustic emission sensor installed on the boiler.

[0127] Specifically, to obtain more detailed and accurate fault information, the system collects current acoustic emission signals from acoustic emission sensors installed on the boiler. These sensors detect elastic waves generated by changes in internal material stress. They can monitor even tiny acoustic emission events caused by localized faults in real time during boiler operation.

[0128] S2054. Extract the frequency spectrum characteristics of the acoustic emission signal.

[0129] Specifically, collected acoustic emission signals are often complex, containing a variety of frequency components. To better analyze these signals, the system requires feature extraction. One common feature extraction method is to extract the signal's spectral characteristics. By converting the acoustic emission signal from the time domain to the frequency domain, the energy distribution of the signal at different frequencies can be more clearly observed. These spectral characteristics can reflect the changes in internal material stress when a fault occurs and are key features for subsequent fault matching and diagnosis.

[0130] S2055: Compare the spectrum characteristics with the known fault characteristic template to obtain a matching result.

[0131] Specifically, after extracting the spectral features of the acoustic emission signal, the system compares them with known fault signature templates. These templates were established through extensive research and experimentation on local boiler faults and contain the spectral features of acoustic emission signals under various typical fault conditions. By comparing the currently collected spectral features with these templates, the system obtains a matching result. The quality of the matching result directly determines the accuracy of the fault diagnosis. A high degree of match indicates that the current acoustic emission signal closely resembles the characteristics of a known fault, allowing a preliminary judgment of the fault type.

[0132] S2056. Determine the fault location based on the matching result.

[0133] Specifically, based on the matching results, the system can determine the specific location where the fault occurred. Due to the propagation characteristics of the acoustic emission signal, by analyzing the signal propagation path and time delay and other information, combined with the structural model of the boiler, the system can accurately locate the location of the fault. This process is crucial for taking effective fault handling measures in a timely manner, because only by accurately knowing the fault location can maintenance personnel quickly arrive at the scene to handle it, thereby minimizing the impact of the fault on the operation of the boiler and ensuring the safe and stable operation of the boiler. The fault tracing module matches the current acoustic emission signal with the preset template through Euclidean distance. The Euclidean distance formula is:

[0134]

[0135] Among them, E current (f) and E template (f) represents the spectrum energy of the current and template at frequency f, respectively. The diagnostic confidence is optimized based on the matching results to improve the accuracy of fault location.

[0136] In an exemplary embodiment, the fault type, location information, and evolution trend are fed back to the boiler control system in real time, triggering a hierarchical early warning strategy and automatically adjusting operating parameters to suppress the development of the fault.

[0137] The boiler control system automatically adjusts the fluidizing air speed and coal feed rate through the PID regulator. The output formula of the PID regulator is:

[0138]

[0139] Among them, K p , K i , K d represents the proportional coefficient, integral coefficient, and differential coefficient, and e(t) represents the deviation between the fault diagnosis result and the target value. Based on the fault diagnosis results, the fluidization air velocity and coal feed rate are adjusted to suppress the fault development and maintain stable boiler operation.

[0140] Based on the same inventive concept, embodiments of the present application also provide a system for implementing the aforementioned method for diagnosing deep peak-shaving faults in circulating fluidized bed boilers. The solution provided by this system is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the circulating fluidized bed boiler deep peak-shaving fault diagnosis system provided below can be found in the aforementioned limitations of the circulating fluidized bed boiler deep peak-shaving fault diagnosis method, and will not be further elaborated here.

[0141] In an exemplary embodiment, Figure 3 As shown, a circulating fluidized bed boiler deep peak regulation fault diagnosis system is provided, comprising:

[0142] An acquisition module 310 is configured to acquire key fault-sensitive features of the circulating fluidized bed boiler when the circulating fluidized bed boiler is in a deep peak-shaving operating condition;

[0143] A construction module 320 is configured to construct a multi-fault coupling analysis model based on the key fault sensitivity characteristics, and obtain a fault probability distribution through the multi-fault coupling analysis model;

[0144] An input module 330 is configured to establish a boiler dynamic characteristics prediction model based on a long short-term memory network, input the fault probability distribution and spatiotemporal coupling characteristics into the boiler dynamic characteristics prediction model, and obtain a three-dimensional diagnosis result;

[0145] A generating module 340 is configured to generate a graded warning signal based on the three-dimensional diagnosis result and an adaptive threshold algorithm;

[0146] The starting module 350 is used to start the local fault tracing module based on transfer learning according to the graded warning signal to achieve fault location.

[0147] As an optional implementation, in the aspect of obtaining key fault-sensitive features of the circulating fluidized bed boiler, the acquisition module 310 is specifically configured to:

[0148] Using a multimodal sensor to collect initial operating parameters of a circulating fluidized bed boiler in real time, the initial operating parameters include boiler load change rate, bed fluidization velocity, and separator pressure difference;

[0149] Normalizing the initial operating parameters;

[0150] Extracting time domain features and frequency domain features from the normalized initial operating parameters, wherein the time domain feature extraction refers to calculating the fluctuation standard deviation and mutation slope of the normalized initial operating parameters, and the frequency domain feature extraction refers to calculating the spectrum energy distribution of the acoustic emission signal of the normalized initial operating parameters;

[0151] Dynamically weighting the time domain features and the frequency domain features to obtain key fault-sensitive features of the circulating fluidized bed boiler.

[0152] As an optional implementation, in terms of dynamically weighting the time domain features and the frequency domain features to obtain the key fault-sensitive features of the circulating fluidized bed boiler, the acquisition module 310 is specifically configured to:

[0153] Calculating a dynamic weighting coefficient according to the normalized boiler load change rate, the normalized bed fluidization velocity, and the normalized separator pressure difference;

[0154] According to the dynamic weighting coefficient, the dynamically weighted fluctuation standard deviation, the dynamically weighted mutation slope and the dynamically weighted frequency domain characteristics are calculated. The dynamically weighted fluctuation standard deviation, the dynamically weighted mutation slope and the dynamically weighted frequency domain characteristics are the key fault-sensitive characteristics of the circulating fluidized bed boiler.

[0155] As an optional implementation, in constructing a multi-fault coupling analysis model based on the key fault-sensitive characteristics of the circulating fluidized bed boiler, the construction module 320 is specifically configured to:

[0156] Based on the fuzzy clustering algorithm, the operating state of the boiler is divided into multiple cluster subspaces, each cluster subspace represents a typical operating mode or a fault precursor state;

[0157] For each of the cluster subspaces, similarity matching is performed between the key fault sensitive features and the fault feature library, and a preliminary fault probability distribution is output based on the matching results;

[0158] The preliminary fault probability distribution is dynamically modified by an adaptive weighted fusion algorithm, and the plurality of cluster subspaces, the preliminary fault probability distribution and the adaptive weighted fusion algorithm constitute the multi-fault coupling analysis model.

[0159] As an optional implementation, in dividing the boiler operating state into multiple cluster subspaces based on the fuzzy clustering algorithm, a module 320 is constructed to specifically:

[0160] Initialize cluster centers and fuzzy coefficients;

[0161] Iteratively calculating the membership of the sample to each of the initialized cluster centers based on the initialized cluster centers and the fuzzy coefficients;

[0162] The initialized cluster centers are updated according to the membership degrees until the cluster centers converge.

[0163] As an optional implementation, in establishing a boiler dynamic characteristics prediction model based on a long short-term memory network, inputting the fault probability distribution and spatiotemporal coupling characteristics into the boiler dynamic characteristics prediction model to obtain a three-dimensional diagnosis result, the input module 330 is specifically used to:

[0164] The standard deviation of oxygen fluctuations is calculated using a sliding window to obtain a combustion stability index in the time dimension. The combustion stability index is used to reflect the stability of the combustion state of the boiler during continuous operation.

[0165] Performing spatial correlation analysis on data collected by the bed pressure sensor array to obtain a bed fluidization uniformity index in the spatial dimension. The bed fluidization uniformity index is used to assess whether there are local anomalies in the coordination of bed material fluidization in different regions.

[0166] Constructing the boiler dynamic characteristics prediction model based on long short-term memory network and attention mechanism;

[0167] The spatiotemporal coupling characteristics and the fault probability distribution are input into the boiler dynamic characteristic prediction model to output a three-dimensional diagnosis result, which includes the fault type, severity and development trend.

[0168] As an optional implementation, the graded warning signal is divided into three levels according to the severity of the fault.

[0169] As an optional implementation, the adaptive threshold algorithm is expressed by the following formula:

[0170]

[0171] Among them, μ history represents the mean of historical fault data, σ history represents the standard deviation of historical fault data, k represents the preset confidence coefficient, and the value range of k is 0.5≤k≤2, N history Indicates the number of historical samples, N current Indicates the current sample size.

[0172] As an optional implementation, in the aspect of starting the local fault tracing module based on transfer learning according to the graded warning signal to achieve fault location, the starting module 350 is specifically configured to:

[0173] judging whether there is a sudden change in bed temperature or abnormal fluidization according to the graded warning signal;

[0174] If the bed temperature suddenly changes or the fluidization is abnormal, activating the local fault tracing module based on transfer learning;

[0175] The acoustic emission signal at the current moment is collected by an acoustic emission sensor installed on the boiler;

[0176] Extracting frequency spectrum characteristics of the acoustic emission signal;

[0177] Comparing the spectrum characteristics with a known fault characteristic template to obtain a matching result;

[0178] The fault location is determined according to the matching result.

[0179] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store key fault-sensitive features. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for diagnosing deep peak-shaving faults of a circulating fluidized bed boiler is implemented.

[0180] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0181] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0182] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0183] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0184] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0185] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0186] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0187] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0188] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. An intelligent diagnosis method for deep peak regulation faults of circulating fluidized bed boilers, characterized in that: The intelligent diagnosis method is used for a circulating fluidized bed boiler. The intelligent diagnosis method for a deep peak-shaving fault of a circulating fluidized bed boiler includes: When the circulating fluidized bed boiler is in deep peak load regulation, key fault sensitive features of the circulating fluidized bed boiler are obtained; Constructing a multi-fault coupling analysis model based on the key fault sensitive characteristics, and obtaining a fault probability distribution through the multi-fault coupling analysis model; Establishing a boiler dynamic characteristics prediction model based on a long short-term memory network, inputting the fault probability distribution and spatiotemporal coupling characteristics into the boiler dynamic characteristics prediction model to obtain a three-dimensional diagnosis result; generating a graded warning signal based on the three-dimensional diagnostic results and an adaptive threshold algorithm; According to the graded warning signal, a local fault tracing module based on transfer learning is started to realize fault location.

2. The intelligent diagnosis method for deep peak-shaving faults of a circulating fluidized bed boiler according to claim 1 is characterized in that: The key fault-sensitive features of the circulating fluidized bed boiler are obtained, including: Using a multimodal sensor to collect initial operating parameters of a circulating fluidized bed boiler in real time, the initial operating parameters include boiler load change rate, bed fluidization velocity, and separator pressure difference; Normalizing the initial operating parameters; Extracting time domain features and frequency domain features from the normalized initial operating parameters, wherein the time domain feature extraction refers to calculating the fluctuation standard deviation and mutation slope of the normalized initial operating parameters, and the frequency domain feature extraction refers to calculating the spectrum energy distribution of the acoustic emission signal of the normalized initial operating parameters; Dynamically weighting the time domain features and the frequency domain features to obtain key fault-sensitive features of the circulating fluidized bed boiler.

3. The intelligent diagnosis method for deep peak-shaving faults of a circulating fluidized bed boiler according to claim 2 is characterized in that: The dynamically weighted processing of the time domain features and the frequency domain features to obtain key fault-sensitive features of the circulating fluidized bed boiler includes: Calculating a dynamic weighting coefficient according to the normalized boiler load change rate, the normalized bed fluidization velocity, and the normalized separator pressure difference; According to the dynamic weighting coefficient, the dynamically weighted fluctuation standard deviation, the dynamically weighted mutation slope and the dynamically weighted frequency domain characteristics are calculated. The dynamically weighted fluctuation standard deviation, the dynamically weighted mutation slope and the dynamically weighted frequency domain characteristics are the key fault-sensitive characteristics of the circulating fluidized bed boiler.

4. The intelligent diagnosis method for deep peak-shaving faults of a circulating fluidized bed boiler according to claim 1, characterized in that: The constructing of a multi-fault coupling analysis model based on the key fault sensitive characteristics includes: Based on the fuzzy clustering algorithm, the operating state of the boiler is divided into multiple cluster subspaces, each cluster subspace represents a typical operating mode or a fault precursor state; For each of the cluster subspaces, similarity matching is performed between the key fault sensitive features and the fault feature library, and a preliminary fault probability distribution is output based on the matching results; The preliminary fault probability distribution is dynamically modified by an adaptive weighted fusion algorithm, and the plurality of cluster subspaces, the preliminary fault probability distribution and the adaptive weighted fusion algorithm constitute the multi-fault coupling analysis model.

5. The intelligent diagnosis method for deep peak-shaving faults of a circulating fluidized bed boiler according to claim 4 is characterized in that: The fuzzy clustering algorithm is used to divide the operating state of the boiler into multiple cluster subspaces, including: Initialize cluster centers and fuzzy coefficients; Iteratively calculating the membership of the sample to each of the initialized cluster centers based on the initialized cluster centers and the fuzzy coefficients; The initialized cluster centers are updated according to the membership degrees until the cluster centers converge.

6. The intelligent diagnosis method for deep peak-shaving faults of a circulating fluidized bed boiler according to claim 1, characterized in that: The boiler dynamic characteristics prediction model is established based on the long short-term memory network, and the fault probability distribution and the spatiotemporal coupling characteristics are input into the boiler dynamic characteristics prediction model to obtain a three-dimensional diagnosis result, including: The standard deviation of oxygen fluctuations is calculated through a sliding window to obtain the combustion stability index in the time dimension; The data collected by the bed pressure sensor array are subjected to spatial correlation analysis to obtain the bed fluidization uniformity index in the spatial dimension; Constructing the boiler dynamic characteristics prediction model based on long short-term memory network and attention mechanism; The spatiotemporal coupling characteristics and the fault probability distribution are input into the boiler dynamic characteristic prediction model to output a three-dimensional diagnosis result, which includes the fault type, severity and development trend.

7. The intelligent diagnosis method for deep peak-shaving faults of a circulating fluidized bed boiler according to claim 1, characterized in that: The adaptive threshold algorithm is dynamically adjusted according to historical fault data and current operating parameters.

8. The intelligent diagnosis method for deep peak regulation fault of circulating fluidized bed boiler according to claim 1, characterized in that: The graded warning signals are divided into three levels according to the severity of the fault.

9. The intelligent diagnosis method for deep peak-shaving faults of a circulating fluidized bed boiler according to claim 1, characterized in that: The adaptive threshold algorithm is expressed by the following formula: Among them, μ history represents the mean of historical fault data, σ history represents the standard deviation of historical fault data, k represents the preset confidence coefficient, and the value range of k is 0.5≤k≤2, N history Indicates the number of historical samples, N current Indicates the current sample size.

10. The intelligent diagnosis method for deep peak regulation fault of circulating fluidized bed boiler according to claim 1, characterized in that: The method of starting a local fault tracing module based on transfer learning according to the graded warning signal to locate the fault includes: judging whether there is a sudden change in bed temperature or abnormal fluidization according to the graded warning signal; If the bed temperature suddenly changes or the fluidization is abnormal, activating the local fault tracing module based on transfer learning; The acoustic emission signal at the current moment is collected by an acoustic emission sensor installed on the boiler; Extracting frequency spectrum characteristics of the acoustic emission signal; Comparing the spectrum characteristics with a known fault characteristic template to obtain a matching result; The fault location is determined according to the matching result.

Images