Deep learning-based boiler wall temperature prediction method and system

Abstract

The invention provides a boiler wall temperature prediction method and system based on deep learning. The method comprises the steps of obtaining characteristic parameters related to boiler wall temperature and prediction parameters during boiler operation; dividing the feature parameters and the prediction parameters into a training set, a verification set and a test set; constructing a boiler wall temperature prediction model; the boiler wall temperature prediction model comprises an empirical mode decomposition module, a deep learning module and an error correction module, the empirical mode decomposition module is used for decomposing prediction parameters into a plurality of component sequences, the deep learning module is used for predicting the boiler wall temperature according to the component sequences, and the error correction module is used for correcting the predicted boiler wall temperature; using the training set to train the boiler wall temperature prediction model, and adjusting the boiler wall temperature prediction model through the verification set; and evaluating the boiler wall temperature prediction model by using the test set, and if an evaluation result meets a preset standard, deploying the boiler wall temperature prediction model into an operation monitoring system of the boiler to predict the boiler wall temperature, thereby realizing accurate prediction of the boiler wall temperature.

Description

Technical Field

[0001] The present invention relates to the field of boiler control technology, and in particular to a boiler wall temperature prediction method and system based on deep learning. Background Art

[0002] Boilers are core equipment in industries such as thermal power generation and chemical production. Accurately predicting their wall temperature is crucial for safe operation and energy efficiency. Boiler walls are exposed to high temperatures and pressures for extended periods of time. Localized overheating can lead to tube bursts, resulting in significant economic losses and safety hazards.

[0003] Traditional wall temperature prediction methods mainly rely on physical model construction or statistical regression analysis based on heat transfer principles. These methods have significant limitations, such as difficulty in accurately describing combustion fluctuations, reliance on expert experience to set characteristic parameters, and real-time prediction accuracy is limited by sensor density and measurement noise interference, making them unable to meet actual production needs. Summary of the Invention

[0004] The present invention provides a boiler wall temperature prediction method and system based on deep learning, which are used to solve the technical problem of insufficient boiler wall temperature prediction accuracy in the prior art.

[0005] In one aspect, the present invention provides a boiler wall temperature prediction method based on deep learning, comprising: Obtain characteristic parameters related to boiler wall temperature and predicted parameters during boiler operation; Dividing the characteristic parameters and the prediction parameters into a training set, a validation set, and a test set; Constructing a boiler wall temperature prediction model; wherein the boiler wall temperature prediction model includes an empirical mode decomposition module, a deep learning module, and an error correction module, wherein the empirical mode decomposition module is used to decompose the prediction parameters into multiple component sequences, the deep learning module is used to predict the boiler wall temperature based on the component sequences, and the error correction module is used to correct the predicted boiler wall temperature; Training the boiler wall temperature prediction model using the training set, and adjusting the boiler wall temperature prediction model using the validation set; The boiler wall temperature prediction model is evaluated using the test set. If the evaluation result meets the preset standard, the boiler wall temperature prediction model is deployed to the boiler operation monitoring system to predict the boiler wall temperature.

[0006] According to a boiler wall temperature prediction method based on deep learning provided by the present invention, the characteristic parameters and the prediction parameters are divided into a training set, a validation set, and a test set, including: According to the diversity of boiler operating conditions, the characteristic parameters and the prediction parameters are divided by stratified sampling to obtain multiple sub-datasets, each sub-dataset representing a different operating condition; Using a sliding window method to divide each of the sub-datasets into a training set, a validation set, and a test set; The division ratio of the training set, the validation set, and the test set is adjusted based on the historical data and real-time data of the boiler operation.

[0007] According to a boiler wall temperature prediction method based on deep learning provided by the present invention, the empirical mode decomposition module is used to decompose the prediction parameters into multiple component sequences, including: The empirical mode decomposition module is used to perform empirical mode decomposition on the time series data of the prediction parameters to obtain multiple eigenmode function components as a component sequence.

[0008] According to a boiler wall temperature prediction method based on deep learning provided by the present invention, the deep learning module is used to predict the boiler wall temperature according to the component sequence, including: The deep learning module uses a bidirectional gated recurrent unit to predict the component sequence to obtain a preliminary prediction value of the boiler wall temperature; wherein an attention mechanism is added to the bidirectional gated recurrent unit to fuse the features in the component sequence; The error correction module is used to correct the predicted boiler wall temperature, including: The error correction module is used to calculate the difference between the preliminary predicted value and the actual value; Analyze the statistical characteristics of the difference; wherein the statistical characteristics include at least one of the mean, variance, and standard deviation; A correction value is generated based on the statistical characteristics, and the preliminary predicted value is corrected to obtain a final boiler wall temperature.

[0009] According to a deep learning-based boiler wall temperature prediction method provided by the present invention, after obtaining characteristic parameters related to the boiler wall temperature and prediction parameters during boiler operation, the method further includes: Performing interpolation processing on missing data or abnormal data in the characteristic parameters and the prediction parameters to supplement the missing data or correct the abnormal data; performing normalization processing on the characteristic parameters and the prediction parameters; A correlation analysis is performed on the characteristic parameters and the prediction parameters, and parts whose correlation with the boiler wall temperature is lower than a preset correlation threshold are removed.

[0010] According to a deep learning-based boiler wall temperature prediction method provided by the present invention, obtaining characteristic parameters related to the boiler wall temperature further includes: By simulating the heat conduction and combustion process inside the boiler, a virtual characteristic parameter related to the wall temperature is generated as a characteristic parameter; and / or, By using the historical data of boiler operation and the time series pattern mining technology, implicit features that are highly correlated with the wall temperature changes are extracted as feature parameters.

[0011] According to a deep learning-based boiler wall temperature prediction method provided by the present invention, after the empirical mode decomposition module is used to perform empirical mode decomposition on the time series data of the prediction parameter to obtain multiple eigenmode function components, the method further includes: Construct the energy-frequency distribution matrix of the eigenmode function components and calculate the coherence coefficient between each component and the wall temperature signal; Based on the coherence coefficient, the screening threshold of the eigenmode function component is constructed; the component screening threshold is shown in the following formula: θ = μ + k·σ; Wherein, μ is the mean of the coherence coefficients of all eigenmode function components, σ ​​is the standard deviation, k is the adaptive adjustment factor, and θ is the component screening threshold; Nonlinearly reorganize the eigenmode function components that satisfy the coherence coefficient greater than the component screening threshold to form a new characteristic component set.

[0012] According to a deep learning-based boiler wall temperature prediction method provided by the present invention, after deploying the boiler wall temperature prediction model into the boiler operation monitoring system, the method further includes: Real-time detection of the boiler's current operating condition characteristic mode; Establish a mapping relationship library between operating condition characteristic patterns and boiler wall temperature prediction model parameters; When a switching of the operating condition is identified, the stopping criterion parameters of the empirical mode decomposition are automatically adjusted according to the mapping relationship library so that the decomposition result adapts to the characteristic scale of the new condition.

[0013] According to a boiler wall temperature prediction method based on deep learning provided by the present invention, the error correction module further includes: A dynamic weight function based on the prediction error change rate is established, where the dynamic weight function is shown in the following formula: W_t=α·(1+tanh(β·Δe_t)); Among them, W_t is the correction weight at the current moment, Δe_t represents the rate of change of the prediction error between time t and time t-1, α is the basic weight coefficient, and β is the sensitivity adjustment coefficient; When the error change is greater than the preset change threshold, the correction weight is increased; when the error change is less than the preset change threshold, the correction weight is decreased; The dynamic weight is integrated with the correction value to form a composite correction strategy.

[0014] On the other hand, the present invention also provides a boiler wall temperature prediction system based on deep learning, comprising: Parameter acquisition module, used to obtain characteristic parameters related to boiler wall temperature and predicted parameters during boiler operation; A parameter division module, used for dividing the characteristic parameters and the prediction parameters into a training set, a validation set and a test set; a model creation module for constructing a boiler wall temperature prediction model; wherein the boiler wall temperature prediction model includes an empirical mode decomposition module, a deep learning module, and an error correction module, wherein the empirical mode decomposition module is used to decompose the prediction parameters into multiple component sequences, the deep learning module is used to predict the boiler wall temperature based on the component sequences, and the error correction module is used to correct the predicted boiler wall temperature; a model training module, configured to train the boiler wall temperature prediction model using the training set and adjust the boiler wall temperature prediction model using the validation set; The model prediction module is used to evaluate the boiler wall temperature prediction model using the test set. If the evaluation result meets the preset standard, the boiler wall temperature prediction model is deployed to the boiler operation monitoring system to predict the boiler wall temperature.

[0015] The deep learning-based boiler wall temperature prediction method and system provided by this invention combines an empirical mode decomposition module, a deep learning module, and an error correction module to construct an efficient boiler wall temperature prediction model. This model effectively handles the nonstationarity and complex dynamic characteristics of boiler wall temperature data. It uses EMD decomposition to reduce signal nonstationarity, captures complex patterns in time series data through deep learning capabilities, and further improves prediction accuracy with the help of an error correction module. Furthermore, through a rational data processing and model training process, the accuracy and reliability of boiler wall temperature predictions can be significantly improved, providing strong support for safe boiler operation and predictive maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 1 is a flow chart of a method for predicting boiler wall temperature based on deep learning provided by an embodiment of the present invention; Figure 2 2 is a schematic structural diagram of a boiler wall temperature prediction system based on deep learning provided by an embodiment of the present invention; Figure 3 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0019] Figure 1 This is a flow chart of a boiler wall temperature prediction method based on deep learning provided by an embodiment of the present invention.

[0020] See also Figure 1 , the boiler wall temperature prediction method based on deep learning may include the following steps 101 to 105.

[0021] Step 101: Obtain characteristic parameters related to boiler wall temperature and predicted parameters during boiler operation.

[0022] In this step, characteristic parameters refer to various factors that can affect the boiler wall temperature, which can include combustion-related parameters (such as coal type and coal quantity, primary air volume, secondary air volume, etc.), working fluid-related parameters (such as feed water flow, main steam pressure, main steam temperature, etc.), and heat exchange-related parameters (such as flue gas outlet temperature, water-cooled wall temperature, superheater temperature and reheater temperature, etc.). These parameters cover all aspects of boiler operation and can provide comprehensive input information for the model.

[0023] Prediction parameters are the target variables to be predicted, specifically the specific values ​​of boiler wall temperatures. These parameters can include water-wall temperature, superheater temperature, and reheater temperature. These parameters are key indicators that require real-time monitoring and prediction during boiler operation.

[0024] Step 102: Divide the feature parameters and prediction parameters into a training set, a validation set, and a test set.

[0025] Step 102 may specifically include: Step 1: Based on the diversity of boiler operating conditions, stratified sampling is used to divide the characteristic parameters and prediction parameters to obtain multiple sub-datasets, each of which represents a different operating condition.

[0026] In step 1, the boiler will be in different operating conditions during operation (such as different loads, combustion conditions, fuel types, etc.), and these operating conditions have different impacts on the wall temperature. Through stratified sampling, we can ensure that each sub-dataset can represent a specific operating condition, so that the model can learn the wall temperature variation patterns under different operating conditions; Based on historical boiler operation data and expert experience, operating conditions are categorized (e.g., high load, low load, normal combustion, abnormal combustion, etc.). Data is then proportionally extracted from each operating condition category to form multiple sub-datasets. Each sub-dataset contains characteristic parameters and predicted parameters for that operating condition.

[0027] Step 2: Use the sliding window method to divide each sub-dataset into training set, validation set and test set.

[0028] In step 2, the boiler wall temperature is time series data, which is dynamic and sequential. The sliding window method can divide time series data into multiple samples, each containing data from a period of time, thereby preserving the temporal characteristics of the data while providing a sufficient number of samples for the model.

[0029] For each sub-dataset, a fixed time window length (e.g., 1 hour or 1 day) is selected and the window is slid along the time series, moving one time step at a time (e.g., 1 minute or 1 hour). At each window position, the data within the window is used as input features, and the wall temperature at the end of the window is used as the label, thereby generating multiple samples, which are divided into training, validation, and test sets.

[0030] Step 3: Combine the historical data and real-time data of boiler operation to adjust the division ratio of training set, validation set and test set.

[0031] In step three, during boiler operation, real-time data and historical data may have different importance. For example, real-time data better reflects current operating conditions, while historical data provides long-term operating trends. By dynamically adjusting the ratio of training, validation, and test sets, we can better balance the training and validation of the model. For example, in the early stages of model development, the proportion of the training set can be increased to improve the model's fit; during the model optimization phase, the proportion of the validation and test sets can be increased to evaluate the model's generalization ability.

[0032] The above steps divide the data through stratified sampling and sliding window methods, and dynamically adjust the division ratio by combining historical and real-time data to ensure the diversity and representativeness of the data during model training, verification and testing. This can effectively improve the model's adaptability and prediction accuracy to different boiler operating conditions, enhance the model's generalization ability, and enable it to more accurately predict boiler wall temperature in practical applications, thereby ensuring the safe operation of the boiler.

[0033] Step 103: Construct a boiler wall temperature prediction model; wherein the boiler wall temperature prediction model includes an empirical mode decomposition module, a deep learning module, and an error correction module, the empirical mode decomposition module is used to decompose the prediction parameters into multiple component sequences, the deep learning module is used to predict the boiler wall temperature based on the component sequences, and the error correction module is used to correct the predicted boiler wall temperature.

[0034] In this step, the empirical mode decomposition module is used to decompose the prediction parameters into multiple component sequences, including: The empirical mode decomposition module is used to perform empirical mode decomposition (EMD) on the time series data of the prediction parameters to obtain multiple intrinsic mode function (IMF) components as a component sequence.

[0035] These component sequences not only reduce the non-stationarity of the signal, but also extract the local features of the signal, providing data that is easier to process and analyze for subsequent deep learning models, significantly improving the model's adaptability to complex working conditions and prediction accuracy.

[0036] Step 104: Use the training set to train the boiler wall temperature prediction model, and adjust the boiler wall temperature prediction model using the validation set.

[0037] In this step, the training set is the dataset used for model learning. It contains the characteristic parameters and the corresponding prediction parameters. By learning the data in the training set, the model finds the mapping between the characteristic parameters and the wall temperature, thus being able to make predictions for new input data.

[0038] The validation set is used to evaluate the model's performance on unseen data and to adjust the model based on the validation results. The validation set data comes from the same distribution as the training set data, but is not used in the model training process. Therefore, it can be used to detect whether the model is overfitting or underfitting.

[0039] During the training process, the model is evaluated using the validation set every certain number of rounds. The specific steps are as follows: The feature parameters in the validation set are input into the model, the predicted output of the model is obtained, and the difference between the predicted value and the actual value is calculated.

[0040] Indicators such as mean square error (MSE) and mean absolute error (MAE) are usually used to evaluate the performance of the model.

[0041] Based on the evaluation results of the validation set, adjust the model's hyperparameters (such as learning rate, number of hidden layer nodes, regularization parameters, etc.). If the model performs poorly on the validation set, you may need to adjust these hyperparameters to improve the model's performance.

[0042] To avoid overfitting, an early stopping mechanism is usually set up, which stops training if the loss of the validation set does not decrease significantly over multiple rounds of training.

[0043] Step 105: Use the test set to evaluate the boiler wall temperature prediction model. If the evaluation result meets the preset standard, deploy the boiler wall temperature prediction model to the boiler operation monitoring system to predict the boiler wall temperature.

[0044] In this example, a highly efficient boiler wall temperature prediction model was constructed by combining an empirical mode decomposition module, a deep learning module, and an error correction module. This model effectively handles the nonstationarity and complex dynamic characteristics of boiler wall temperature data. EMD decomposition is used to reduce signal nonstationarity, while deep learning captures complex patterns in time series data. The error correction module further improves prediction accuracy. Furthermore, through a rational data processing and model training process, the accuracy and reliability of boiler wall temperature predictions can be significantly improved, providing strong support for safe boiler operation and predictive maintenance.

[0045] In one embodiment of the present specification, the deep learning module is used to predict the boiler wall temperature based on the component sequence, including: The deep learning module uses a bidirectional gated recurrent unit (Bi-GRU) to predict the component sequences and obtain a preliminary prediction of the boiler wall temperature. The Bi-GRU incorporates an attention mechanism to fuse features from the component sequences. The error correction module is used to correct the predicted boiler wall temperature, including: The error correction module is used to calculate the difference between the preliminary predicted value and the actual value; Analyze statistical characteristics of the difference; wherein the statistical characteristics include at least one of mean, variance, and standard deviation; A correction value is generated based on the statistical characteristics, and the preliminary prediction value is corrected to obtain the final boiler wall temperature.

[0046] In this example, compared to the unidirectional GRU, the Bi-GRU can utilize more contextual information and reduce prediction errors. Furthermore, the GRU's structure is simpler than the LSTM, with fewer parameters (approximately 30% less). This allows for faster training when processing high-frequency time series data and a stronger ability to filter short-term fluctuations, avoiding the overfitting problem that can occur with the complex gating mechanism of the LSTM.

[0047] Based on the output of Bi-GRU, the attention mechanism is introduced. The attention mechanism calculates the importance weights of different components at each time step and performs weighted summation on the output of Bi-GRU to generate the final preliminary prediction value. This weighted fusion method can highlight important features, suppress unimportant features, and further improve prediction accuracy.

[0048] Correction values ​​are generated based on statistical characteristics (such as mean and variance). For example, if the mean of the error is positive, it indicates that the predicted values ​​are generally low, and a correction value can be added to adjust the prediction results. If the variance of the error is large, it indicates that the error fluctuates greatly, and the correction value can be adjusted to smooth the error. Finally, the correction value is added to the initial prediction value to obtain the final prediction result.

[0049] In one embodiment of the present specification, after obtaining characteristic parameters related to the boiler wall temperature and predicted parameters during boiler operation, the method further includes: Interpolate missing data or abnormal data in feature parameters and prediction parameters to supplement missing data or correct abnormal data; Normalize the feature parameters and prediction parameters; Correlation analysis is performed on the characteristic parameters and prediction parameters, and the parts with correlation with boiler wall temperature lower than the preset correlation threshold are removed.

[0050] In this embodiment, common interpolation methods include linear interpolation, polynomial interpolation, spline interpolation, etc. For example, if the water-cooled wall temperature data at a certain point in time is missing, it can be filled by linear interpolation of normal data at adjacent time points. Outliers can be identified by statistical methods (such as mean, median, standard deviation) or anomaly detection algorithms based on machine learning. Once an outlier is detected, it can be replaced by the average value of the adjacent normal data or other reasonable values. Calculate the correlation coefficient between each characteristic parameter and the boiler wall temperature (such as Pearson correlation coefficient, Spearman rank correlation coefficient, etc.). The value range of the correlation coefficient is usually between [-1,1]. The closer the absolute value is to 1, the stronger the correlation. Set a correlation threshold (such as 0.5 or 0.6) according to actual needs. For characteristic parameters with correlation coefficients lower than the threshold, it is considered that their relationship with the wall temperature is weak and they can be removed from the model input.

[0051] In one embodiment of this specification, obtaining characteristic parameters related to boiler wall temperature also includes: By simulating the heat conduction and combustion process inside the boiler, a virtual characteristic parameter related to the wall temperature is generated as a characteristic parameter; and / or, By using the historical data of boiler operation and the time series pattern mining technology, implicit features that are highly correlated with the wall temperature changes are extracted as feature parameters.

[0052] In this embodiment, boiler wall temperature is affected by a variety of complex physical processes, such as heat release during combustion and heat transfer during heat conduction. These processes may be difficult to measure directly with sensors or may be prohibitively expensive. Heat transfer within the boiler wall is simulated using heat conduction equations (such as Fourier's law of heat conduction). The temperature distribution at different locations within the wall is calculated based on parameters such as wall thickness, material thermal conductivity, and internal and external surface temperatures, thereby generating virtual characteristic parameters related to the wall temperature. Combustion processes within the boiler are simulated using combustion models (such as chemical reaction kinetics models) to calculate the heat generated by combustion and the distribution of combustion products. These simulation results serve as virtual characteristic parameters to reflect the impact of the combustion process on the wall temperature. These virtual characteristic parameters can complement actual measurement data, providing a more comprehensive description of physical processes, thereby enhancing the model's understanding of boiler operating conditions and improving prediction accuracy.

[0053] A large amount of historical data accumulates during boiler operation. This data contains a wealth of information, including patterns in wall temperature variation and characteristic patterns under different operating conditions. Time series pattern mining techniques can be used to extract implicit features from this historical data that are valuable for wall temperature prediction. Specifically, historical time series data of boiler wall temperature are analyzed to extract its statistical characteristics (such as mean, variance, and autocorrelation coefficient) and trend characteristics (such as seasonal variations and long-term trends). Machine learning or data mining algorithms (such as cluster analysis and association rule mining) are then used to mine patterns or features from this historical data that are highly correlated with wall temperature variations. For example, cluster analysis can reveal patterns of wall temperature variation under different operating conditions, while association rule mining can identify feature combinations that are relevant to wall temperature variations.

[0054] Suppose there is a set of historical data on boiler operation, including features such as wall temperature, coal quantity, primary air volume, and secondary air volume. By analyzing this data, we can extract the following implicit features. The mean and variance of wall temperature: reflect the average level and degree of fluctuation of wall temperature; the correlation between coal quantity and wall temperature: it is found that the wall temperature increases when the coal quantity increases; the long-term trend and seasonal changes of wall temperature: it is found that the wall temperature changes regularly during specific time periods of each day; the results of operating condition clustering: the operating conditions are divided into high load and low load categories, each corresponding to a different wall temperature change pattern; the abnormal pattern of wall temperature: it is found that a sudden and significant drop in wall temperature may be due to abnormal combustion. These implicit features can be used as input to the model to help the model better understand the operating status of the boiler, thereby improving the accuracy of wall temperature prediction.

[0055] In one embodiment of the present specification, after the empirical mode decomposition module is used to perform empirical mode decomposition (EMD) on the time series data of the prediction parameter to obtain multiple intrinsic mode function (IMF) components, the method further includes: Step 1: Construct the energy-frequency distribution matrix of the intrinsic mode function (IMF) components and calculate the coherence coefficient between each component and the wall temperature signal; In this step, empirical mode decomposition (EMD) decomposes the complex non-stationary time series into multiple intrinsic mode function (IMF) components. Each IMF component represents the characteristics of different frequencies and energies in the original signal. By analyzing the energy and frequency distribution of these components, we can better understand their contribution to the wall temperature signal.

[0056] Specifically, for each IMF component, the corresponding energy can be obtained by integrating the square of the component. The instantaneous frequency of each component can be obtained through the Hilbert transform. The energy and frequency information of each IMF component are integrated into a matrix to form an energy-frequency distribution matrix. Each row of the matrix corresponds to an IMF component, and each column represents the energy and frequency, respectively.

[0057] Use Fourier transform or other correlation analysis methods to calculate the coherence coefficient between each IMF component and the wall temperature signal. The coherence coefficient ranges from [0 to 1], with values ​​closer to 1 indicating stronger correlation.

[0058] Step 2: Based on the coherence coefficient, construct the screening threshold of the intrinsic mode function (IMF) component; the component screening threshold is shown in the following formula (1): θ = μ + k·σ (1); Wherein, μ is the mean of the coherence coefficients of all intrinsic mode functions (IMF) components, σ ​​is the standard deviation, k is the adaptive adjustment factor, and θ is the component screening threshold; Step 3: Perform nonlinear reorganization on the intrinsic mode function (IMF) components that satisfy the coherence coefficient greater than the component screening threshold to form a new characteristic component set.

[0059] In this step, the selected IMF components are nonlinearly combined, for example, by weighted summation, multiplication, or other nonlinear functions. The purpose of nonlinear recombination is to further extract and fuse the features of these components to form a new set of feature components.

[0060] In this embodiment, the components obtained by empirical mode decomposition are optimized by constructing an energy-frequency distribution matrix, calculating the coherence coefficient, designing a screening threshold, and nonlinearly reorganizing the IMF components. This can effectively screen out components that have important contributions to the wall temperature prediction, and further optimize the feature set through nonlinear reorganization, thereby improving the prediction performance and efficiency of the model.

[0061] In one embodiment of the present specification, after the boiler wall temperature prediction model is deployed in the boiler operation monitoring system, the following steps are further included: Step 1: Real-time detection of the boiler's current operating condition characteristic mode; During operation, the boiler switches to different operating conditions based on load, fuel type, and combustion conditions. These changes can affect the wall temperature prediction. Therefore, real-time detection of the current operating condition's characteristic patterns is crucial for ensuring model adaptability. Specifically, key features, such as load, fuel type, and combustion parameters, are extracted from real-time boiler operating data. Machine learning algorithms (such as cluster analysis and support vector machines) are then used to classify these features and identify the current operating condition. For example, the operating conditions can be categorized as high load, low load, normal combustion, or abnormal combustion.

[0062] Step 2: Establish a mapping relationship library between operating condition characteristic patterns and boiler wall temperature prediction model parameters; In this step, different operating conditions have different effects on the wall temperature, so different model parameters need to be configured for each operating condition. The mapping relationship library is used to store the optimal model parameters corresponding to each operating condition.

[0063] Specifically, during the model training phase, model parameters are optimized for each operating condition. For example, cross-validation is used to find the most suitable parameters for high-load conditions and low-load conditions. Each operating condition and its corresponding optimal model parameters are stored in a mapping library. The mapping library can be a database or data table that records the correspondence between the characteristic patterns of the operating conditions and the model parameters.

[0064] Step 3: When the operating condition is identified to be switched, the stopping criterion parameters of the empirical mode decomposition (EMD) decomposition are automatically adjusted according to the mapping relationship library so that the decomposition result adapts to the characteristic scale of the new operating condition.

[0065] In this step, empirical mode decomposition (EMD) is an adaptive signal processing method whose decomposition results are influenced by stopping criterion parameters. Different operating conditions may require different stopping criterion parameters to ensure that the decomposition results accurately reflect the characteristics of the current operating condition. Specifically, changes in the characteristic patterns of the operating conditions are monitored in real time. When a switching operating condition is detected, the model parameter adjustment is triggered. Parameter adjustment automatically adjusts the EMD decomposition parameters based on the optimal stopping criterion parameters for the corresponding operating condition stored in the mapping relationship library. For example, the number of EMD iterations and the screening threshold are adjusted.

[0066] In this embodiment, by real-time detection of operating condition characteristic patterns, establishment of a mapping relationship library, and dynamic adjustment of EMD decomposition parameters, the boiler wall temperature prediction model is adaptively adjusted under different operating conditions, ensuring that the model can adapt quickly when the operating conditions change and maintain a high prediction accuracy, thereby providing more reliable protection for the safe operation and predictive maintenance of the boiler.

[0067] In one embodiment of the present specification, the error correction module further includes: A dynamic weight function based on the prediction error change rate is established, where the dynamic weight function is shown in the following formula (2): W_t=α·(1+tanh(β·Δe_t))(2); Among them, W_t is the correction weight at the current moment, Δe_t represents the rate of change of the prediction error between time t and time t-1, α is the basic weight coefficient, and β is the sensitivity adjustment coefficient; When the error change is greater than the preset change threshold, the correction weight is increased; when the error change is less than the preset change threshold, the correction weight is decreased; The dynamic weights are integrated with the correction values ​​to form a composite correction strategy.

[0068] In this embodiment, a flexible and effective error correction strategy is formed by establishing a dynamic weight function, adjusting the correction weight based on the error change rate, and fusing the dynamic weight with the correction value. This strategy can dynamically adjust the correction strength based on the error change, thereby maintaining high prediction accuracy under different circumstances and further optimizing the performance of the boiler wall temperature prediction model. The fusion method can generally be summation or weighted summation.

[0069] In some other embodiments of this specification, the boiler wall temperature prediction method based on deep learning further includes: Create a "multi-scale collaborative prediction" architecture: divide boiler wall temperature prediction into two scales: macro trend prediction and micro fluctuation prediction, and improve overall prediction accuracy through collaborative optimization; specifically, it includes: Macro trend prediction: Using historical data on boiler operation, a long-term prediction sub-model is trained to capture the overall trend of wall temperature changes; Micro-fluctuation prediction: Based on high-frequency sampling data, a short-period prediction sub-model is trained to capture the instantaneous fluctuation characteristics of wall temperature; Collaborative optimization: By combining macro and micro prediction results through a dynamic fusion algorithm and adjusting the fusion weights based on prediction error feedback, collaborative optimization of multi-scale predictions is achieved, thereby improving the sensitivity of short-term predictions while ensuring long-term stability.

[0070] Based on the same general inventive concept, the present invention also protects a boiler wall temperature prediction system based on deep learning, such as Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of a deep learning-based boiler wall temperature prediction system provided by an embodiment of the present invention. The deep learning-based boiler wall temperature prediction system provided by the present invention is described below. The deep learning-based boiler wall temperature prediction system described below and the deep learning-based boiler wall temperature prediction method described above can be used in conjunction with each other.

[0071] The boiler wall temperature prediction system based on deep learning includes a parameter acquisition module 201, a parameter division module 202, a model creation module 203, a model training module 204 and a model prediction module 205.

[0072] The parameter acquisition module 201 is used to obtain characteristic parameters related to the boiler wall temperature and predicted parameters during boiler operation; The parameter division module 202 is used to divide the feature parameters and the prediction parameters into a training set, a validation set and a test set; The model creation module 203 is used to construct a boiler wall temperature prediction model; wherein the boiler wall temperature prediction model includes an empirical mode decomposition module, a deep learning module, and an error correction module, wherein the empirical mode decomposition module is used to decompose the prediction parameters into multiple component sequences, the deep learning module is used to predict the boiler wall temperature based on the component sequences, and the error correction module is used to correct the predicted boiler wall temperature; The model training module 204 is used to train the boiler wall temperature prediction model using the training set and adjust the boiler wall temperature prediction model using the validation set; The model prediction module 205 is used to evaluate the boiler wall temperature prediction model using the test set. If the evaluation result meets the preset standard, the boiler wall temperature prediction model is deployed to the boiler operation monitoring system to predict the boiler wall temperature.

[0073] Figure 3 It is a structural diagram of an electronic device provided by an embodiment of the present invention.

[0074] like Figure 3 As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340. The processor 310, the communication interface 320, and the memory 330 communicate with each other via the communication bus 340. The processor 310 may call logic instructions in the memory 330 to execute the boiler wall temperature prediction method based on deep learning.

[0075] Furthermore, the logic instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0076] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the boiler wall temperature prediction method based on deep learning provided by the above methods.

[0077] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the boiler wall temperature prediction method based on deep learning provided by the above methods.

[0078] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0079] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A boiler wall temperature prediction method based on deep learning, characterized in that: include: Obtain characteristic parameters related to boiler wall temperature and predicted parameters during boiler operation; Dividing the characteristic parameters and the prediction parameters into a training set, a validation set, and a test set; Constructing a boiler wall temperature prediction model; wherein the boiler wall temperature prediction model includes an empirical mode decomposition module, a deep learning module, and an error correction module, wherein the empirical mode decomposition module is used to decompose the prediction parameters into multiple component sequences, the deep learning module is used to predict the boiler wall temperature based on the component sequences, and the error correction module is used to correct the predicted boiler wall temperature; Training the boiler wall temperature prediction model using the training set, and adjusting the boiler wall temperature prediction model using the validation set; The boiler wall temperature prediction model is evaluated using the test set. If the evaluation result meets the preset standard, the boiler wall temperature prediction model is deployed to the boiler operation monitoring system to predict the boiler wall temperature.

2. The boiler wall temperature prediction method based on deep learning according to claim 1 is characterized in that: The dividing the characteristic parameters and the prediction parameters into a training set, a validation set, and a test set comprises: According to the diversity of boiler operating conditions, the characteristic parameters and the prediction parameters are divided by stratified sampling to obtain multiple sub-datasets, each sub-dataset representing a different operating condition; Using a sliding window method to divide each of the sub-datasets into a training set, a validation set, and a test set; The division ratio of the training set, the validation set, and the test set is adjusted based on the historical data and real-time data of the boiler operation.

3. The boiler wall temperature prediction method based on deep learning according to claim 1, characterized in that: The empirical mode decomposition module is used to decompose the prediction parameters into multiple component sequences, including: The empirical mode decomposition module is used to perform empirical mode decomposition on the time series data of the prediction parameters to obtain multiple eigenmode function components as a component sequence.

4. The boiler wall temperature prediction method based on deep learning according to claim 3 is characterized in that: The deep learning module is used to predict the boiler wall temperature based on the component sequence, including: The deep learning module uses a bidirectional gated recurrent unit to predict the component sequence to obtain a preliminary prediction value of the boiler wall temperature; wherein an attention mechanism is added to the bidirectional gated recurrent unit to fuse the features in the component sequence; The error correction module is used to correct the predicted boiler wall temperature, including: The error correction module is used to calculate the difference between the preliminary predicted value and the actual value; Analyze the statistical characteristics of the difference; wherein the statistical characteristics include at least one of the mean, variance, and standard deviation; A correction value is generated based on the statistical characteristics, and the preliminary predicted value is corrected to obtain a final boiler wall temperature.

5. The boiler wall temperature prediction method based on deep learning according to claim 1, characterized in that: After obtaining characteristic parameters related to the boiler wall temperature and predicted parameters during boiler operation, the method further includes: Performing interpolation processing on missing data or abnormal data in the characteristic parameters and the prediction parameters to supplement the missing data or correct the abnormal data; performing normalization processing on the characteristic parameters and the prediction parameters; A correlation analysis is performed on the characteristic parameters and the prediction parameters, and parts whose correlation with the boiler wall temperature is lower than a preset correlation threshold are removed.

6. The boiler wall temperature prediction method based on deep learning according to claim 1, characterized in that: The obtaining of characteristic parameters related to the boiler wall temperature also includes: By simulating the heat conduction and combustion process inside the boiler, a virtual characteristic parameter related to the wall temperature is generated as a characteristic parameter; and / or, By using the historical data of boiler operation and the time series pattern mining technology, implicit features that are highly correlated with the wall temperature changes are extracted as feature parameters.

7. The boiler wall temperature prediction method based on deep learning according to claim 3 is characterized in that: After the empirical mode decomposition module is used to perform empirical mode decomposition on the time series data of the prediction parameter to obtain a plurality of eigenmode function components, the method further includes: Construct the energy-frequency distribution matrix of the eigenmode function components and calculate the coherence coefficient between each component and the wall temperature signal; Based on the coherence coefficient, the screening threshold of the eigenmode function component is constructed; the component screening threshold is shown in the following formula: θ = μ + k·σ; Wherein, μ is the mean of the coherence coefficients of all eigenmode function components, σ ​​is the standard deviation, k is the adaptive adjustment factor, and θ is the component screening threshold; Nonlinearly reorganize the eigenmode function components that satisfy the coherence coefficient greater than the component screening threshold to form a new characteristic component set.

8. The boiler wall temperature prediction method based on deep learning according to claim 3 is characterized in that: After the boiler wall temperature prediction model is deployed in the boiler operation monitoring system, the method further includes: Real-time detection of the boiler's current operating condition characteristic mode; Establish a mapping relationship library between operating condition characteristic patterns and boiler wall temperature prediction model parameters; When a switching of the operating condition is identified, the stopping criterion parameters of the empirical mode decomposition are automatically adjusted according to the mapping relationship library so that the decomposition result adapts to the characteristic scale of the new condition.

9. The boiler wall temperature prediction method based on deep learning according to claim 4, characterized in that: The error correction module further includes: A dynamic weight function based on the prediction error change rate is established, where the dynamic weight function is shown in the following formula: W_t=α·(1+tanh(β·Δe_t)); Among them, W_t is the correction weight at the current moment, Δe_t represents the rate of change of the prediction error between time t and time t-1, α is the basic weight coefficient, and β is the sensitivity adjustment coefficient; When the error change is greater than the preset change threshold, the correction weight is increased; when the error change is less than the preset change threshold, the correction weight is decreased; The dynamic weight is integrated with the correction value to form a composite correction strategy.

10. A boiler wall temperature prediction system based on deep learning, characterized in that: include: Parameter acquisition module, used to obtain characteristic parameters related to boiler wall temperature and predicted parameters during boiler operation; A parameter division module, used for dividing the characteristic parameters and the prediction parameters into a training set, a validation set and a test set; a model creation module for constructing a boiler wall temperature prediction model; wherein the boiler wall temperature prediction model includes an empirical mode decomposition module, a deep learning module, and an error correction module, wherein the empirical mode decomposition module is used to decompose the prediction parameters into multiple component sequences, the deep learning module is used to predict the boiler wall temperature based on the component sequences, and the error correction module is used to correct the predicted boiler wall temperature; a model training module, configured to train the boiler wall temperature prediction model using the training set and adjust the boiler wall temperature prediction model using the validation set; The model prediction module is used to evaluate the boiler wall temperature prediction model using the test set. If the evaluation result meets the preset standard, the boiler wall temperature prediction model is deployed to the boiler operation monitoring system to predict the boiler wall temperature.

Images