Boiler furnace combustion state detection method, system, medium and equipment

Abstract

The invention provides a boiler hearth combustion state detection method. The method comprises the steps that historical operation data of a boiler are obtained; extracting dynamic embedded features, and combining the dynamic embedded features with original features corresponding to the operation data to obtain boiler hearth combustion feature data; constructing a combustion state detection model based on the boiler furnace combustion characteristic data; variational inference is conducted on the constructed combustion state detection model, model parameters are obtained through optimization, and an optimal combustion state detection model is obtained; and acquiring real-time observation data of the boiler, inputting the real-time observation data into the optimal combustion state detection model, outputting a state probability and a corresponding confidence interval, and determining a hearth combustion state of the boiler. The real-time performance and the model interpretability of the combustion state detection model are improved, the judgment precision of the boiler combustion state is improved, and the risk of the boiler production process is effectively reduced. The invention further provides a boiler hearth combustion state detection system, a computer readable storage medium and electronic equipment which have the above beneficial effects.

Description

Technical Field

[0001] The present application relates to the field of industrial control systems, and in particular to a method, system, medium and equipment for detecting the combustion state of a boiler furnace. Background Art

[0002] Trending failures in process systems occur when equipment or system performance gradually deteriorates over long-term operation due to physical, chemical, or environmental factors, ultimately leading to functional failure. These failures develop slowly, and initial symptoms are difficult to detect, often only becoming noticeable once the failure is more severe. Trending failures are common in continuously operating chemical plants. Timely monitoring and detection of trending failures are crucial for reducing energy consumption, optimizing production efficiency, and minimizing unplanned downtime.

[0003] Existing fault diagnosis methods, such as static model-based Hidden Markov Model (HMM) or analysis of single sensor signals, usually assume that system state changes are sudden and their prediction accuracy is limited.

[0004] Furthermore, the nonlinear evolution of trending faults increases modeling complexity, making conventional methods ineffective when dealing with time dependencies and dynamic coupling. Furthermore, industrial sites demand high real-time performance, requiring diagnostic systems to respond quickly. Traditional methods often rely on offline analysis and struggle to adapt to new data immediately, leading to delayed fault warnings and increased downtime and maintenance costs. Summary of the Invention

[0005] The purpose of this application is to provide a boiler furnace combustion state detection method, system, computer-readable storage medium and electronic equipment, which can improve the accuracy of boiler furnace combustion state judgment and reduce generation risk.

[0006] To solve the above technical problems, the present application provides a method for detecting the combustion state of a boiler furnace. The specific technical solution is as follows:

[0007] Obtain historical operating data of the boiler; the operating data includes observation data of the slag cooler speed, oxygen content and material bed differential pressure when the boiler is in normal operating state and abnormal operating state;

[0008] Extracting dynamic embedded features of the historical operating data, and combining the dynamic embedded features with original features corresponding to the operating data to obtain boiler furnace combustion feature data;

[0009] Building a combustion state detection model based on the boiler furnace combustion characteristic data;

[0010] performing variational inference on the constructed combustion state detection model to optimize the obtained model parameters, and applying the model parameters to the combustion state detection model to obtain an optimal combustion state detection model;

[0011] Obtaining real-time observation data of the boiler, inputting it into the optimal combustion state detection model, and outputting the state probability and the corresponding confidence interval;

[0012] The furnace combustion state of the boiler is determined according to the state probability and the confidence interval.

[0013] Optionally, extracting dynamic embedding features of the historical operation data includes:

[0014] constructing a Hankel matrix based on the historical operating data;

[0015] Singular value decomposition is performed on the Hankel matrix, and dynamic embedding features are obtained according to the influence ratio of the principal components.

[0016] Optionally, constructing a combustion state detection model based on the boiler furnace combustion characteristic data includes:

[0017] A variational inference hidden Markov model is used as the model architecture of the combustion state detection model, and the model hidden state, transition probability, and emission probability are defined; the model hidden state is used to characterize the fault stage of the boiler; the transition probability is used to describe the dynamic conversion law between different fault stages; and the emission probability is used to characterize the mapping relationship between the model hidden state and the observed data;

[0018] A combustion state detection model is constructed according to the boiler furnace combustion characteristic data, the model hidden state, the transition probability and the emission probability.

[0019] Optionally, the process of defining the transition probability includes:

[0020] A dynamic function corresponding to the transfer probability is defined according to the control transfer trend parameter, trend change time, pressure impact trend fault variable and flow impact trend fault; the dynamic function includes the boiler time accumulation effect, slag cooler speed change and oxygen content change.

[0021] Optionally, performing variational inference on the constructed combustion state detection model to optimize the model parameters includes:

[0022] Initialize the state sequence distribution corresponding to the approximate distribution of variational inference, the transfer parameter Gaussian distribution, the mean Gaussian distribution, and the covariance inverse Wishart distribution;

[0023] Taking the evidence lower bound as the objective function, calculating the gradient of the evidence lower bound with respect to the variational parameter;

[0024] The variational parameters are iteratively updated using a gradient ascent algorithm to serve as model parameters when the objective function is maximized.

[0025] Optionally, after obtaining the real-time observation data of the boiler, the following is also included:

[0026] Calculating a data distribution distance between the optimal combustion state detection model and the real-time observation data;

[0027] If the data distribution distance exceeds a preset value, obtain the data weight;

[0028] A weighted evidence lower bound is calculated based on the data weight, and the weighted evidence lower bound is applied to replace the evidence lower bound to update the variational parameter.

[0029] The present application also provides a boiler furnace combustion state detection system, comprising:

[0030] A historical data acquisition module is used to acquire historical operating data of the boiler; the operating data includes observation data of the slag cooler speed, oxygen content and material bed differential pressure when the boiler is in normal operating state and abnormal operating state respectively;

[0031] a feature extraction module, configured to extract dynamic embedded features of the historical operating data, and combine the dynamic embedded features with original features corresponding to the operating data to obtain boiler furnace combustion feature data;

[0032] A model building module, configured to build a combustion state detection model based on the boiler furnace combustion characteristic data;

[0033] a model parameter inference module, configured to perform variational inference on the constructed combustion state detection model, optimize the obtained model parameters, and apply the model parameters to the combustion state detection model to obtain an optimal combustion state detection model;

[0034] The state detection module is used to obtain real-time observation data of the boiler, input it into the optimal combustion state detection model, output the state probability and the corresponding confidence interval; and determine the furnace combustion state of the boiler based on the state probability and the confidence interval.

[0035] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the boiler furnace combustion state detection method as described above are implemented.

[0036] The present application also provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the steps of the boiler furnace combustion state detection method as described above are implemented.

[0037] The present application also provides a computer program product, comprising a computer program, which implements the steps of the boiler furnace combustion state detection method as described above when the computer program is executed.

[0038] The present application provides a method for detecting the combustion state of a boiler furnace, comprising: obtaining historical operating data of the boiler; the operating data comprising observation data of a slag cooler speed, oxygen content, and differential pressure of a material bed when the boiler is in a normal operating state and an abnormal operating state; extracting dynamic embedding features of the historical operating data, combining the dynamic embedding features with original features corresponding to the operating data, and obtaining boiler furnace combustion characteristic data; constructing a combustion state detection model based on the boiler furnace combustion characteristic data; performing variational inference on the constructed combustion state detection model, optimizing model parameters, and applying the model parameters to the combustion state detection model to obtain an optimal combustion state detection model; obtaining real-time observation data of the boiler, inputting the data into the optimal combustion state detection model, and outputting a state probability and a corresponding confidence interval; and determining the furnace combustion state of the boiler based on the state probability and the confidence interval.

[0039] After obtaining the historical operating data of the boiler, this application enhances the ability to capture the details of the boiler's operating characteristics by extracting dynamic embedded features, and at the same time constructs a combustion state detection model, and uses variational inference to perform online optimization of the combustion state detection model to ensure that the combustion state detection model has a high real-time state judgment capability, improves the real-time performance and model interpretability of the combustion state detection model, improves the accuracy of the judgment of the boiler's combustion state, and effectively reduces the risk of the boiler production process.

[0040] The present application also provides a boiler furnace combustion state detection system, a computer-readable storage medium and an electronic device, which have the above-mentioned beneficial effects and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] 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 or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0042] Figure 1 A flow chart of a method for detecting the combustion state of a boiler furnace provided in an embodiment of the present application;

[0043] Figure 2 A schematic diagram of the structure of a boiler furnace combustion state detection system provided in an embodiment of the present application;

[0044] Figure 3This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are 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.

[0046] See also Figure 1 , Figure 1 This is a flow chart of a method for detecting the combustion state of a boiler furnace provided in an embodiment of the present application. The method includes:

[0047] S101: Acquire historical operating data of the boiler; the operating data includes observation data of the slag cooler speed, oxygen content, and material bed differential pressure when the boiler is in normal operating state and abnormal operating state respectively;

[0048] S102: extracting dynamic embedded features of the historical operating data, and combining the dynamic embedded features with original features corresponding to the operating data to obtain boiler furnace combustion feature data;

[0049] S103: Building a combustion state detection model based on the boiler furnace combustion characteristic data;

[0050] S104: performing variational inference on the constructed combustion state detection model to optimize model parameters, and applying the model parameters to the combustion state detection model to obtain an optimal combustion state detection model;

[0051] S105: Acquire real-time observation data of the boiler, input it into the optimal combustion state detection model, and output the state probability and the corresponding confidence interval;

[0052] S106: Determine the furnace combustion state of the boiler according to the state probability and the confidence interval.

[0053] In step S101, observation data of the slag cooler speed, oxygen content, and bed pressure difference during normal operation and multi-stage failures, i.e., historical operation data, can be collected from the boiler combustion process:

[0054] .

[0055] At the same time, corresponding data preprocessing can be performed on the acquired historical operation data, which is not specifically limited here.

[0056] Specifically, the process history data set can be collected from the process sensors , where the number of variables is m, expressed as. Take the sampling point , and the outliers of individual variables are proposed and the z scores are normalized. The results are expressed as .

[0057] Use historical operation data or pre-processed historical operation data to perform dynamic feature extraction to obtain dynamic embedding features , and Combined with the original observed historical operation data, the boiler furnace combustion characteristic data is obtained:

[0058] ;

[0059] Specifically, a Hankel matrix can be constructed based on the historical operation data, and then singular value decomposition can be performed on the Hankel matrix to output dynamic embedding features based on the influence ratio of the principal components. In this case, the following steps can be specifically included:

[0060] The first step is to use the data to construct the Hankel matrix;

[0061] Step 2: Singular Value Decomposition (SVD);

[0062] Step 3: Output the adjusted observed variables for diagnosis .

[0063] When using data to construct the Hankel matrix, use the historical operating data obtained Constructing the Hankel matrix :

[0064] ;

[0065] in is the delay dimension, which indicates the embedding time window size.

[0066] Hankel matrix Perform singular value decomposition calculation:

[0067] ;

[0068] Check the decay of singular values ​​and retain the top values ​​based on the influence ratio of the principal component (e.g. 90%) Principal components. Finally, dynamic embedding features are extracted:

[0069] ;

[0070] Observed variables used for diagnostics after output adjustment , the dynamic embedding feature Combined with the original observed historical operation data, the boiler furnace combustion characteristic data is obtained:

[0071] ;

[0072] Thereafter, in step S103, based on the boiler furnace combustion characteristic data To build a combustion state detection model for the fault state. There is no limitation on how to build the combustion state detection model. This embodiment uses the VI-HMM model as an example to illustrate the model construction and variational inference process:

[0073] For the VI-HMM model, the variational inference hidden Markov model is used as the model architecture of the combustion state detection model, and the model hidden state, transition probability and emission probability are defined; the model hidden state is used to characterize the fault stage of the boiler; the transition probability is used to describe the dynamic conversion law between different fault stages; the emission probability is used to characterize the mapping relationship between the model hidden state and the observation data

[0074] The model hidden state refers to the self-defined system internal failure stage (such as normal, thin layer thickness, very thick layer thickness), which is the diagnosis target. The model hidden state can be written as:

[0075] ;

[0076] The specific number of hidden states Can be tuned based on experience and is usually a discrete value.

[0077] The transition probability is the dynamic transition law between states, describing how the progressive behavior evolves with time and conditions. becomes The transition probability is recorded as The transition probability is defined as a dynamic function that is time- and observation-dependent:

[0078] .

[0079] in, To control the parameters of the transfer trend, is the time of trend change, are the variables that affect trend failure, which are pressure and flow respectively.

[0080] Transfer probability does not include bed pressure difference This is mainly because it is assumed in the design that the state transition of the furnace material layer thickness is mainly affected by the cumulative effect of time (t), the change of the slag cooler speed ( ) and oxygen content changes ( ) is driven by the bed pressure difference The influence on the change of furnace layer thickness is relatively minor or indirect.

[0081] The emission probability represents the mapping between the state and the observation data, connecting the internal state and the external measurement, denoted as , can be assumed to be normally distributed:

[0082] ;

[0083] Combined with historical operating data , hidden state and parameters , the joint distribution of the VI-HMM model can be expressed as:

[0084] ;

[0085] In step S104, the distribution is approximated according to the variational method. The state sequence distribution, transfer parameter Gaussian distribution, mean Gaussian distribution and covariance inverse Wishart distribution corresponding to the variational inference approximate distribution can be initialized, and the evidence lower bound is used as the objective function. The gradient of the evidence lower bound with respect to the variational parameter is calculated, and finally the variational parameter is iteratively updated using the gradient ascent algorithm to serve as the model parameter when the objective function is maximized.

[0086] Specifically, by introducing the approximate distribution , according to mean field theory, assuming factorization:

[0087] ;

[0088] in, is the state sequence distribution, is the transfer parameter Gaussian distribution, is a Gaussian distribution with mean, is the covariance inverse Wishart distribution.

[0089] The objective function is defined as the evidence lower bound, and VI approximates the true posterior by maximizing the evidence lower bound.

[0090] ;

[0091] in:

[0092] ;

[0093] The first term is the expectation of the joint distribution, and the second term is the entropy of the approximate distribution. (KL divergence is non-negative), so maximizing the evidence lower bound is equivalent to minimizing KL.

[0094] In a feasible implementation, the uniform distribution can be initialized first. , zero mean and unit covariance , data mean ,as well as ,Right now .

[0095] Update first :

[0096] Forward algorithm:

[0097] ;

[0098] Backward algorithm:

[0099] ;

[0100] State probability:

[0101] ;

[0102] renew , the gradient of the lower bound of the evidence is:

[0103] ;

[0104] The update method is:

[0105] ;

[0106] in A custom learning step size.

[0107] Updated later , The mean can be expressed as:

[0108] ;

[0109] ;

[0110] The covariance is:

[0111] ;

[0112] .

[0113] When calculating the lower bound of evidence, if the change is less than , stop the iteration and output the optimal combustion state detection model at this time, otherwise repeat the above steps to continue updating the parameters.

[0114] Based on the above embodiment, a feasible implementation method is to further limit the update weight of fault data when updating the combustion state detection model in real time, considering the problem that the baseline model may be excessively affected by faults and gradually "drift". The specific update steps are as follows:

[0115] The first step is to calculate the data distribution distance between the optimal combustion state detection model and the real-time observation data;

[0116] Step 2: If the data distribution distance exceeds a preset value, obtain the data weight;

[0117] The third step is to calculate a weighted evidence lower bound based on the data weight, and apply the weighted evidence lower bound to replace the evidence lower bound to update the variational parameter.

[0118] While there is no limitation on the data distribution distance used, one feasible implementation is the Mahalanobis distance. The Mahalanobis distance is a calculation method that measures the distance between two points in multidimensional space, taking into account the covariance structure of the data. The following is the process for calculating the Mahalanobis distance between the optimal combustion state detection model and real-time observation data:

[0119] ;

[0120] like (three standard deviations), it is marked as a potential failure.

[0121] Defining data weights , you can refer to the following formula:

[0122] ;

[0123] in is a custom parameter. The weight of normal data , fault data .

[0124] In the update of the weighted evidence lower bound, first modify the single-step evidence lower bound:

[0125] ;

[0126] Then, when the gradient is updated, it is combined with the data weight:

[0127] .

[0128] Furthermore, when obtaining the real-time observation data of the boiler, inputting it into the optimal combustion state detection model, and outputting the state probability and the corresponding confidence interval, the state probability and uncertainty are used to perform multi-level warning. , the uncertainty is:

[0129] .

[0130] Finally, according to the state probability Provide early warning information and determine the current fault status. Approximately normal distribution, according to the uncertainty, the 95% confidence interval is:

[0131] ;

[0132] Ultimately, the boiler furnace combustion state can be determined based on the state probability and confidence interval. The state probability measures the likelihood of the boiler furnace combustion being in different states (such as normal combustion, unstable combustion, and deflagration). For example, a state probability of 0.8 indicates an 80% probability of being in a particular combustion state. The confidence interval provides a credible range for the state probability estimate. For example, for a state probability of 0.8, the 95% confidence interval might be [0.75, 0.85]. This indicates that in repeated combustion state assessments, there is a 95% probability that the true state probability will fall within this interval.

[0133] This embodiment does not limit the classification of boiler furnace combustion states; however, they can generally be categorized as normal combustion, unstable combustion, flameout, and deflagration. The boiler furnace combustion state is determined based on the calculated state probability and confidence interval. If the probability of a particular state is high (e.g., greater than 0.9) and the confidence interval is narrow (e.g., less than 0.1), the boiler furnace can be determined with a high degree of certainty to be in that state. For example, if the calculated probability of a normal combustion state is 0.95 and the confidence interval is [0.93, 0.97], the boiler furnace can be determined to be in a normal combustion state. Conversely, if the state probability is low (e.g., less than 0.5) or the confidence interval is wide (e.g., greater than 0.2), the combustion state is uncertain, and further data collection or verification of data quality and model rationality is required.

[0134] After obtaining the historical operating data of the boiler, the embodiment of the present application enhances the ability to capture details of the boiler operating characteristics by extracting dynamic embedded features, and at the same time constructs a combustion state detection model, and uses variational inference to perform online optimization of the combustion state detection model to ensure that the combustion state detection model has a high real-time state judgment capability, improves the real-time performance and model interpretability of the combustion state detection model, improves the judgment accuracy of the boiler combustion state, and effectively reduces the risk of the boiler production process.

[0135] On this basis, the application of a variational inference hidden Markov model (VI-HMM) achieves real-time state estimation and uncertainty quantification through probabilistic modeling and online optimization. This overcomes the limitations of traditional methods, such as insufficient dynamic modeling, poor real-time performance, and weak interpretability. It provides an efficient and accurate diagnostic tool for processes with trending faults, effectively reducing the risk of production interruptions and optimizing maintenance strategies. This embodiment achieves real-time diagnosis of trending faults under finite variables by extracting dynamic embedded features and employing a variational inference hidden Markov model, overcoming the limitations of traditional methods in dynamic modeling, real-time performance, and data sparsity. This facilitates the implementation of appropriate early warnings after determining the combustion state of the boiler furnace, further enhancing the model's adaptability and reliability, and providing an efficient and practical solution for trending fault diagnosis.

[0136] In order to better understand the boiler furnace combustion state detection method provided by the present application, the boiler furnace combustion state detection process is described below using the actual application calculation process of the present application:

[0137] Consider a boiler furnace combustion process where the variable collected is the slag cooler speed , oxygen content , bed pressure difference .

[0138] First, historical operating data is obtained from the boiler furnace combustion, including the historical data of the slag cooler speed, oxygen content, and bed differential pressure under various operating conditions (speed r / min, oxygen content mol / m 3 , bed differential pressure Pa), where the data operating conditions include normal operating data and operating data for three different discrete material layer thickness levels. In this example, expert experience and industry standards are used to preliminarily define the mean values ​​of normal data and the three levels of material layer thickness as follows:

[0139] normal:

[0140] .

[0141] Thinner layer thickness:

[0142] .

[0143] Thick layer thickness:

[0144] .

[0145] Very thick layer thickness:

[0146] .

[0147] 10,000 consecutive samples were collected for each data. Furthermore, the z-score normalization method was used to preprocess the data, and the obtained ; .

[0148] Using the data of different operating conditions contained in the historical operating data, HAVOK calculation is performed to extract the respective dynamic embedding features Considering the number of variables is 3, the time embedding window size is , then the Hankel matrix formed by each operating condition is for:

[0149] ;

[0150] Then perform SVD singular value decomposition on each Hankel matrix:

[0151] ;

[0152] ;

[0153] in ; ; . Further, truncated SVD, select the front principal components, for example, the first three principal components that account for 90% of the singular value energy can be selected , extract the dynamic embeddings of four working conditions:

[0154] ;

[0155] ;

[0156] The extracted dynamic embeddings are then combined with the original data to form enhanced observations.

[0157] .

[0158] ;

[0159] ;

[0160] ;

[0161] ;

[0162] And similarly, this enhanced observation is established for each set of data.

[0163] Define the hidden states of normal data and three levels of fault data (normal, thin layer thickness, thick layer thickness, very thick layer thickness), the corresponding statistical means and variances are expressed as normal : , ;Thin layer thickness : , Thicker material layer thickness : , Very thick layer thickness : , .

[0164] The functional form of the transition probability is:

[0165] .

[0166] Assuming the initialization , ; ; ; ; .

[0167] The emission function has the form:

[0168] .

[0169] Assume that the initial mean and variance are:

[0170] .

[0171] .

[0172] .

[0173] .

[0174] Variational distribution parameters and initialization:

[0175] :

[0176] :

[0177] .

[0178] :

[0179] .

[0180] :

[0181] .

[0182] :

[0183] .

[0184] :

[0185] .

[0186] .

[0187] :

[0188] .

[0189] .

[0190] :

[0191] .

[0192] .

[0193] The variational EM algorithm is used to iteratively optimize the lower bound of evidence.

[0194] renew , calculated using the forward-backward algorithm .

[0195] Forward:

[0196] .

[0197]

[0198] For example, , .

[0199] .

[0200] Backward:

[0201] .

[0202] renew:

[0203] .

[0204] For example ,

[0205] ; ; .

[0206] Step M: Update .

[0207] :

[0208] For example, ;

[0209] .

[0210] .

[0211] :

[0212] .

[0213] For example, .

[0214] .

[0215] .

[0216] .

[0217] :

[0218] .

[0219] .

[0220] For example, ; .

[0221] .

[0222] .

[0223] Repeat the E and M steps until the lower bound of evidence converges. The formula for the lower bound of evidence is:

[0224] ;

[0225] Starting from the first term, they are initial state prior, transition probability expectation, emission probability expectation, parameter prior, state entropy, and parameter entropy.

[0226] Initial state prior :

[0227] Assumptions ,

[0228] .

[0229] expected transition probability :

[0230] by , , For example:

[0231] ,

[0232] ,

[0233] .

[0234] (Thinner layer thickness):

[0235] .

[0236] .

[0237] .

[0238] .

[0239] (Thicker layer thickness):

[0240] .

[0241] .

[0242] .

[0243] (Thickness of very thick layer):

[0244] .

[0245] sum:

[0246] Normal segment (9999 times): .

[0247] Thinner material layer thickness section: .

[0248] Thicker material layer thickness section: .

[0249] Very thick layer thickness section: .

[0250] total: .

[0251] The expected emission probability :

[0252] set up (normal), , .

[0253] .

[0254] ; .

[0255] ;

[0256] ;

[0257] ; ; .

[0258] .

[0259] (thinner layer thickness), , .

[0260] , .

[0261] , .

[0262] .

[0263] (Thicker layer thickness), , , .

[0264] (very thick layer thickness), , ,

[0265] .

[0266] sum:

[0267] Normal segment: 10000 .

[0268] Thinner layer thickness section: 10000 .

[0269] Thicker material layer thickness: 10000 .

[0270] Very thick layer thickness section: .

[0271] total: .

[0272] Parameter priors :

[0273] ;

[0274] Assumptions: .

[0275] :

[0276] .

[0277] .

[0278] .

[0279] A total of 16 , assuming an average: .

[0280] :

[0281] .

[0282] .

[0283] .

[0284] 4 states: .

[0285] :

[0286] Assumptions (simplify).

[0287] 4: ,

[0288] Sum of parameter priors: .

[0289] State entropy :

[0290] by For example,

[0291] .

[0292] total: .

[0293] Parameter entropy :

[0294] .

[0295] :

[0296] .

[0297] 16: .

[0298] :

[0299] .

[0300] 4: .

[0301] :

[0302] Assumptions .

[0303] 4: .

[0304] total:

[0305] Aggregate evidence lower bound:

[0306] .

[0307] After training, assume that the normal state parameters are:

[0308] ;

[0309] ;

[0310] The status parameters for other states are:

[0311] ; .

[0312] ; .

[0313] ; .

[0314] Consider Start, update every minute. Sliding window size , HAVOK parameters .

[0315] Update the HAVOK embedding using a sliding window. First, collect samples and , the window is (100 samples), and normalized, then constructed the Hankel matrix H, performed SVD, extracted .

[0316] Raw data:

[0317] .

[0318] .

[0319] .

[0320] Normalization:

[0321] .

[0322] .

[0323] .

[0324] Arrange into Hankel matrix , after SVD, merge with the original data, for example, when , the enhanced data obtained .

[0325] Then incrementally optimize the VI-HMM parameters. Calculate the lower bound of the single-step evidence and update , .

[0326] .

[0327] The results of the example calculation are as follows:

[0328] Initial state:

[0329] .

[0330] Emission probability:

[0331] .

[0332] .

[0333] .

[0334] :

[0335] .

[0336] ;

[0337] :

[0338] .

[0339] ;

[0340] :

[0341] .

[0342] .

[0343] Transition probability:

[0344] ;

[0345] .

[0346] .

[0347] .

[0348] renew :

[0349] .

[0350] .

[0351] .

[0352] .

[0353] Parameter updates, such as updating :

[0354] .

[0355] .

[0356] Adapting to new data:

[0357] Will Input the updated VI-HMM, infer the state and record it.

[0358] , indicating a trend towards thicker layer thickness.

[0359] Fine-tune to .

[0360] Weighted updates limit the impact of faulty data:

[0361] Calculate Mahalanobis distance and determine weight .

[0362] Weighted Update:

[0363] .

[0364] The example calculation process is as follows, Mahalanobis distance:

[0365] .

[0366] .

[0367] .

[0368] Weight:

[0369] .

[0370] Weighted Update:

[0371] .

[0372] .

[0373] Assume the current time is , the optimal combustion state detection model has been trained and updated online, and the normal state parameters are:

[0374] ; .

[0375] ; .

[0376] ; .

[0377] ; .

[0378] The enhanced data is:

[0379] .

[0380] Warning rules:

[0381] Thinner layer thickness: .

[0382] (The probability is high, the material layer starts to thicken).

[0383] Very thick layer thickness: (High probability of very thick layer thickness).

[0384] Uncertainty: Using state distribution entropy .

[0385] like , high uncertainty.

[0386] Calculate the current state probability:

[0387] Using the online updated optimal combustion state detection model, based on and the previous moment state calculation . Combine the forward algorithm and transition probability to update the state distribution.

[0388] Example calculation:

[0389] Emission probability:

[0390] .

[0391] .

[0392] .

[0393] .

[0394] .

[0395] Transition probability:

[0396] .

[0397] .

[0398] .

[0399] ,

[0400] Forward algorithm:

[0401] .

[0402] .

[0403] .

[0404] .

[0405] Normalization:

[0406] sum: .

[0407] .

[0408] .

[0409] .

[0410] .

[0411] .

[0412] Determine the thickness of the material layer:

[0413] according to Determine the status.

[0414] Example:

[0415] (The probability is lowest for very thick layers).

[0416] (The probability of thicker material layer thickness is second).

[0417] (The probability is lower for thinner layer thickness).

[0418] (The probability of normal state is the smallest).

[0419] Judgment: Tends to have a very thick material layer.

[0420] Uncertainty calculation:

[0421] entropy:

[0422] .

[0423] , indicating high uncertainty and requiring careful judgment.

[0424] determination:

[0425] although Highest, but not reaching the very thick layer thickness threshold .

[0426] , does not meet the thicker material layer thickness warning.

[0427] , does not trigger the thinner material layer thickness warning.

[0428] High Entropy This indicates that the state distribution is dispersed and may be a transitional stage.

[0429] Output alarm information:

[0430] Status: The current status of the material layer thickness is:

[0431] normal: .

[0432] Thinner layer thickness: .

[0433] Thick layer thickness: .

[0434] Very thick layer thickness: .

[0435] Judgment: The material layer may be in the transition stage from moderate to very thick layer thickness, but has not reached the warning threshold.

[0436] Uncertainty: The state entropy is , uncertainty is high, and continuous monitoring is recommended.

[0437] At this time, a voice alarm can be executed, and the alarm content can be as follows:

[0438] "Warning: The risk of material layer thickness increases and may be approaching moderate or very thick material layer thickness. Please closely monitor the slag cooler speed, oxygen content and material bed differential pressure data."

[0439] It can be seen from the above specific examples that the present application can combine dynamic embedding feature extraction with variational inference hidden Markov model, extract dynamic features from finite variables through Hankel matrix and singular value decomposition and combine them with variational hidden Markov model to achieve accurate modeling of trend faults.

[0440] By designing dynamic transition probabilities and constructing a state transition function based on time and real-time boiler operating data, we can capture the gradual evolution of trending faults. This application also includes an online variational inference mechanism, utilizing a sliding window to update dynamic embedded features and single-step evidence lower bound optimization to ensure real-time state inference.

[0441] See also Figure 2 , Figure 2 A schematic diagram of a boiler furnace combustion state detection system provided in an embodiment of the present application is provided. The system includes:

[0442] A historical data acquisition module is used to acquire historical operating data of the boiler; the operating data includes observation data of the slag cooler speed, oxygen content and material bed differential pressure when the boiler is in normal operating state and abnormal operating state respectively;

[0443] a feature extraction module, configured to extract dynamic embedded features of the historical operating data, and combine the dynamic embedded features with original features corresponding to the operating data to obtain boiler furnace combustion feature data;

[0444] A model building module, configured to build a combustion state detection model based on the boiler furnace combustion characteristic data;

[0445] a model parameter inference module, configured to perform variational inference on the constructed combustion state detection model, optimize the obtained model parameters, and apply the model parameters to the combustion state detection model to obtain an optimal combustion state detection model;

[0446] The state detection module is used to obtain real-time observation data of the boiler, input it into the optimal combustion state detection model, output the state probability and the corresponding confidence interval; and determine the furnace combustion state of the boiler based on the state probability and the confidence interval.

[0447] The present application also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method described in the above method embodiment.

[0448] It is understandable that if the method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and executes all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0449] The computer-readable storage medium provided in this embodiment includes the above-mentioned method, and the effect is the same as above.

[0450] The present application also provides a computer program product, including a computer program, which, when executed, implements the steps corresponding to the above-mentioned boiler furnace combustion state detection method embodiment.

[0451] This application also provides an electronic device, see Figure 3 , a structural diagram of an electronic device provided in an embodiment of the present application, such as Figure 3 As shown, a processor 1410 and a memory 1420 may be included.

[0452] The processor 1410 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1410 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 1410 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 1410 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1410 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0453] The memory 1420 may include one or more computer-readable storage media, which may be non-transitory. The memory 1420 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 1420 is at least used to store the following computer program 1421, wherein, after the computer program is loaded and executed by the processor 1410, it can implement the relevant steps in the method performed by the electronic device side disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 1420 may also include an operating system 1422 and data 1423, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 1422 may include Windows, Linux, Android, etc.

[0454] In some embodiments, the electronic device may further include a display screen 1430 , an input / output interface 1440 , a communication interface 1450 , a sensor 1460 , a power supply 1470 , and a communication bus 1480 .

[0455] certainly, Figure 3 The structure of the electronic device shown does not constitute a limitation on the electronic device in the embodiment of the present application. In actual applications, the electronic device may include Figure 3 More or fewer components than shown, or combinations of certain components.

[0456] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems provided in the embodiments, since they correspond to the methods provided in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0457] 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 ideas of this application. It should be noted that for those skilled in the art, without departing from the principles of this application, various improvements and modifications can be made to this application, and such improvements and modifications also fall within the scope of protection of this application.

[0458] It should also be noted that, in this specification, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

Claims

1. A method for detecting the combustion state of a boiler furnace, characterized in that: include: Obtain historical operating data of the boiler; the operating data includes observation data of the slag cooler speed, oxygen content and material bed differential pressure when the boiler is in normal operating state and abnormal operating state; Extracting dynamic embedded features of the historical operating data, and combining the dynamic embedded features with original features corresponding to the operating data to obtain boiler furnace combustion feature data; Building a combustion state detection model based on the boiler furnace combustion characteristic data; performing variational inference on the constructed combustion state detection model to optimize the obtained model parameters, and applying the model parameters to the combustion state detection model to obtain an optimal combustion state detection model; Obtaining real-time observation data of the boiler, inputting it into the optimal combustion state detection model, and outputting the state probability and the corresponding confidence interval; The furnace combustion state of the boiler is determined according to the state probability and the confidence interval.

2. The method for detecting the combustion state of a boiler furnace according to claim 1, characterized in that: The extracting of dynamic embedding features of the historical operation data includes: constructing a Hankel matrix based on the historical operating data; Singular value decomposition is performed on the Hankel matrix, and dynamic embedding features are obtained according to the influence ratio of the principal components.

3. The method for detecting the combustion state of a boiler furnace according to claim 1, wherein: Constructing a combustion state detection model based on the boiler furnace combustion characteristic data includes: A variational inference hidden Markov model is used as the model architecture of the combustion state detection model, and the model hidden state, transition probability, and emission probability are defined; the model hidden state is used to characterize the fault stage of the boiler; the transition probability is used to describe the dynamic conversion law between different fault stages; and the emission probability is used to characterize the mapping relationship between the model hidden state and the observed data; A combustion state detection model is constructed according to the boiler furnace combustion characteristic data, the model hidden state, the transition probability and the emission probability.

4. The method for detecting the combustion state of a boiler furnace according to claim 3, characterized in that: The process of defining the transition probability includes: A dynamic function corresponding to the transfer probability is defined according to the control transfer trend parameter, trend change time, pressure impact trend fault variable and flow impact trend fault; the dynamic function includes the boiler time accumulation effect, slag cooler speed change and oxygen content change.

5. The method for detecting the combustion state of a boiler furnace according to claim 1, characterized in that: Variational inference is performed on the combustion state detection model, and the model parameters obtained by optimization include: Initialize the state sequence distribution corresponding to the approximate distribution of the variational inference, the transfer parameter Gaussian distribution, the mean Gaussian distribution, and the covariance inverse Wishart distribution; Taking the evidence lower bound as the objective function, calculating the gradient of the evidence lower bound with respect to the variational parameter; The variational parameters are iteratively updated using a gradient ascent algorithm to serve as model parameters when the objective function is maximized.

6. The method for detecting the combustion state of a boiler furnace according to claim 5, characterized in that: After obtaining the real-time observation data of the boiler, it also includes: Calculating a data distribution distance between the optimal combustion state detection model and the real-time observation data; If the data distribution distance exceeds a preset value, obtain the data weight; A weighted evidence lower bound is calculated based on the data weight, and the weighted evidence lower bound is applied to replace the evidence lower bound to update the variational parameter.

7. A boiler furnace combustion state detection system, characterized in that: include: A historical data acquisition module is used to acquire historical operating data of the boiler; the operating data includes observation data of the slag cooler speed, oxygen content and material bed differential pressure when the boiler is in normal operating state and abnormal operating state respectively; a feature extraction module, configured to extract dynamic embedded features of the historical operating data, and combine the dynamic embedded features with original features corresponding to the operating data to obtain boiler furnace combustion feature data; A model building module, configured to build a combustion state detection model based on the boiler furnace combustion characteristic data; a model parameter inference module, configured to perform variational inference on the constructed combustion state detection model, optimize the obtained model parameters, and apply the model parameters to the combustion state detection model to obtain an optimal combustion state detection model; The state detection module is used to obtain real-time observation data of the boiler, input it into the optimal combustion state detection model, output the state probability and the corresponding confidence interval; and determine the furnace combustion state of the boiler based on the state probability and the confidence interval.

8. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the method according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which implements the steps of the method according to any one of claims 1 to 6 when executed.

10. A computer program product, characterized in that The invention comprises a computer program, which implements the steps of the method according to any one of claims 1 to 6 when the computer program is executed.

Images