Boiler four-tube leakage early warning method and system

Abstract

The present invention relates to the technical field of thermal power generation, and in particular to a boiler four-tube leakage early warning method and system, wherein the method comprises the following steps: S1: acquiring power plant operating parameters and dissolved hydrogen data of the boiler four-tube; S2: storing the power plant operating parameters and the dissolved hydrogen data in a time series database through a data middleware; S3: selecting fault characteristic parameters related to the boiler four-tube leakage from the power plant operating parameters and performing data preprocessing; S4: determining an optimal neural network prediction model based on the fault characteristic parameters; S5: acquiring boiler parameters and establishing a tube wall temperature prediction model for the boiler four-tube; S6: establishing an oxide film thickness calculation model, and establishing a corrosion monitoring model based on the oxide film thickness calculation model, the tube wall temperature prediction model and the dissolved hydrogen data; S7: constructing a fusion model based on the optimal neural network prediction model and the corrosion monitoring model, and performing a boiler four-tube leakage early warning based on the fusion value. The method has higher accuracy and reliability.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal power generation, and in particular to a boiler four-tube leakage early warning method and system. Background Art

[0002] A thermal power plant, also known as a thermal power plant, is a facility that uses combustible materials as fuel to produce electricity. A thermal power plant includes fuel systems, combustion systems, steam-water systems, electrical systems, and control systems. Key components of these systems are boilers, steam turbines, and generators, which are housed within the power plant's main building.

[0003] Once a thermal power plant is turned on, it operates continuously. The economizer, water-wall, superheater, and reheater tubes in a boiler are collectively referred to as the "four tubes." They fulfill the crucial functions of heating, evaporation, and transmission. When boilers operate under high temperature and pressure for extended periods, leaks within these four tubes are common.

[0004] In the related art, the leakage warning of the four boiler pipes mainly adopts the acoustic detection method based on signal processing technology and the radiographic detection method based on ray detection. In the acoustic detection method based on signal processing technology, the detection results of the acoustic detection device are highly dependent on factors such as the quantity and quality of process data, the deployment location of the acoustic detection device, the interference of background noise, and the attenuation of sound waves. In the radiographic detection method based on ray detection, the radiographic detection device can detect pores, cracks, and oxide layers that are undetectable to the naked eye. Therefore, the deployment location and number of the radiographic detection devices, the refraction angle of the rays, the attenuation coefficient, and the imaging technology will all affect the leakage detection results. It can be seen that the above two methods cannot detect boiler pipe leaks in advance, and can only detect the abnormal sound of "bursting pipes" after the leak occurs.

[0005] Because the four pipes in a boiler are interconnected, their operating parameters are highly coupled. This structure often leads to similarities in the symptoms of pipeline leaks. In this context, big data-based neural network technology offers a natural advantage in detecting four-pipe boiler leaks. However, relying solely on data processing can lead to discrepancies between the results and the actual situation. This is especially true given the diverse operating conditions of coal-fired boilers. Even if the same heating surface leaks under different operating conditions, the corresponding operating parameters may vary. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a boiler four-tube leakage early warning method and system. The above method and system use a dissolved hydrogen detection device and deep learning to combine to issue an early warning for boiler four-tube leakage, which has higher accuracy and reliability than the existing technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a boiler four-tube leakage early warning method, comprising the following steps:

[0009] S1: Obtain power plant operating parameters and obtain dissolved hydrogen data of the four boiler tubes based on the dissolved hydrogen detection device;

[0010] S2: Build database middleware and store power plant operating parameters and dissolved hydrogen data in a time series database through the data middleware;

[0011] S3: Select fault characteristic parameters related to boiler four-tube leakage from the power plant operation parameters, and perform data preprocessing on the fault characteristic parameters;

[0012] S4: Determine the optimal neural network prediction model based on the fault characteristic parameters after data preprocessing;

[0013] S5: Obtain boiler parameters and establish a wall temperature prediction model for the four boiler tubes based on the boiler parameters;

[0014] S6: Establish an oxide film thickness calculation model based on the pipe wall temperature prediction model, and establish a corrosion monitoring model based on the oxide film thickness calculation model, the pipe wall temperature prediction model, and the dissolved hydrogen data;

[0015] S7: A fusion model is constructed based on the optimal neural network prediction model and the corrosion monitoring model. The multi-source redundant information is weighted averaged based on the fusion model to obtain a fusion value. The boiler four-tube leakage warning is performed based on the fusion value.

[0016] By adopting the above technical solution, when providing a leak warning for the four boiler tubes, power plant operating parameters can be directly obtained, and dissolved hydrogen data in the water vapor of the four boiler tubes can be obtained through a dissolved hydrogen detection device. The power plant operating parameters and dissolved hydrogen data can then be stored in a time-series database through the constructed data middleware for easy access. The power plant operating parameters are then filtered to obtain fault characteristic parameters related to the four boiler tubes' leaks. The fault characteristic parameters are then processed to obtain the optimal neural network prediction model for fault identification. Furthermore, the boiler parameters can be used to establish a wall temperature prediction model for the four boiler tubes. This wall temperature prediction model can be used to obtain the wall temperature of the four boiler tubes. Based on the wall temperature, an oxide film thickness calculation model can be determined. This oxide film thickness calculation model can then be used to obtain the oxide film thickness in the four boiler tubes. Using the wall temperature, oxide film thickness, and dissolved hydrogen data, a corrosion monitoring model for the four boiler tubes can be established.

[0017] Finally, the optimal neural network prediction model is fused with the corrosion monitoring model to create a fusion model. This model performs a weighted average of the leakage warning information output by the optimal neural network prediction model and the leakage warning information output by the corrosion monitoring model to obtain a fusion value. By analyzing this fusion value, we can determine the leakage status of the four boiler pipes and provide a leak warning for the four boiler pipes.

[0018] Optionally, step S3 includes the following steps:

[0019] S31: selecting fault characteristic parameters related to boiler four-tube leakage from power plant operation parameters;

[0020] S32: performing data cleaning on the fault characteristic parameters to obtain a cleaned data set;

[0021] S33: Perform data dimensionality reduction on the cleaned data set to obtain low-dimensional feature parameters;

[0022] S34: Divide the low-dimensional feature parameters into training set, validation set and test set.

[0023] By adopting this technical solution, the power plant operating parameters, including those of the power plant's boilers, steam turbines, and generators, are selected to obtain fault characteristic parameters associated with four-tube boiler leakage. These fault characteristic parameters are then preprocessed and subjected to data dimensionality reduction to obtain low-dimensional feature parameters. This data reduction can preserve the original information structure or maximize specific features. Finally, the low-dimensional feature parameters are normalized and divided into training, validation, and test sets for subsequent steps.

[0024] Optionally, the fault characteristic parameters include: actual load, main steam pressure, feed water flow, induced draft impeller opening, left high-temperature superheater outlet steam temperature, right high-temperature superheater outlet steam temperature, left high-temperature superheater outlet steam pressure, right high-temperature superheater outlet steam pressure, left superheater secondary cooling water flow, right superheater secondary cooling water flow, left high-temperature reheater outlet steam temperature, right high-temperature reheater outlet steam temperature, left high-temperature reheater outlet steam pressure, right high-temperature reheater outlet steam pressure, left high-temperature reheater cooling water flow, right reheater cooling water flow, left and right economizer outlet flue gas temperature difference, economizer outlet feed water temperature, water-cooled wall outlet header temperature, and steam-water separator outlet steam pressure.

[0025] By adopting the above technical solution and obtaining the above fault characteristic parameters, data related to the leakage conditions of the four boiler tubes can be obtained.

[0026] Optionally, step S4 includes the following steps:

[0027] S41: Construct a CNN-LSTM neural network model based on the training set, validation set, convolutional neural network model, and long short-term memory network model;

[0028] S42: Determine the structural parameters of the optimal neural network prediction model based on the validation set, test set and CNN-LSTM neural network model.

[0029] By adopting the above technical solution, a CNN-LSTM neural network model can be constructed through a convolutional neural network model and a long short-term memory network model. The CNN-LSTM neural network model can be trained through a training set, and the CNN-LSTM neural network model can be tested through a test set.

[0030] Optionally, step S5 includes the following steps:

[0031] S51: Obtain boiler parameters and establish a combustion numerical simulation model based on the boiler parameters;

[0032] The boiler parameters include fuel characteristics, fuel flow, furnace size, furnace temperature, pressure, and flue gas output. The boiler parameters are divided using an unstructured grid. The boundary conditions of the combustion numerical simulation model include primary air volume and temperature, air volume and temperature around the primary air nozzle, exhaust gas volume and temperature, air volume and temperature around the exhaust gas nozzle, secondary air volume distribution and secondary air temperature, and burnout air volume and temperature.

[0033] S52: Establish a wall temperature prediction model for the four boiler tubes based on the combustion numerical simulation model.

[0034] By adopting the above technical solution, a combustion numerical simulation model can be established through boiler parameters, and then a tube wall temperature prediction model of the four tubes of the boiler can be established through the combustion numerical simulation model.

[0035] Optionally, the combustion numerical simulation model is verified by comparing the simulation values ​​and design values ​​of the average flue gas temperatures at the inlet and outlet of the boiler water-cooled wall tubes, superheater tubes, reheater tubes, and economizer tubes under the boiler maximum output condition, boiler rated load condition, heat rate acceptance condition, 40%THA, 50%THA, and 75%THA conditions.

[0036] By adopting the above technical solution, the combustion numerical simulation model can be verified by comparing the simulation values ​​and design values ​​of the average flue gas temperatures at the inlet and outlet of the boiler water-cooled wall tubes, superheater tubes, reheater tubes, and economizer tubes under the boiler maximum output condition, boiler rated load condition, heat rate acceptance condition, 40%THA, 50%THA, and 75%THA conditions.

[0037] Optionally, step S6 includes the following steps:

[0038] S61: Determine the outer wall temperature of the four boiler tubes based on the tube wall temperature prediction model and the power plant operation parameters;

[0039] S62: Establish an oxide film thickness calculation model based on the outer wall temperature of the four boiler tubes;

[0040] S63: A corrosion monitoring model is established based on the oxide film thickness calculation model, the pipe wall temperature prediction model, and the dissolved hydrogen data.

[0041] By adopting the above technical solution, the outer wall temperature of the four boiler tubes can be determined by inputting the power plant operating parameters into the tube wall temperature prediction model. The oxide film thickness calculation model can be established based on the outer wall temperature. Finally, the corrosion monitoring model can be established based on the post-oxidation film, tube wall temperature and dissolved hydrogen data.

[0042] Optionally, the boiler four-tube leakage early warning method further includes the following steps:

[0043] S8: Build a cockpit for boiler four-tube leakage warning data and display the boiler four-tube leakage warning results.

[0044] By adopting the above technical solution, the cockpit can display the leakage warning results of the four boiler tubes.

[0045] In a second aspect, the present invention provides a boiler four-tube leakage warning system, comprising a dissolved hydrogen detection device and a processing device, wherein the dissolved hydrogen detection device is connected to the economizer tube, the water-cooled wall tube, the superheater tube and the reheater tube, and the processing device is communicatively connected to the dissolved hydrogen detection device. The processing device includes a processor, a memory and a communication bus, wherein the communication bus is used to realize the communication connection between the processor and the memory, and the processor is used to execute a computer program stored in the memory to realize the boiler four-tube leakage warning method as described in any one of the first aspects.

[0046] By adopting the above technical solution, the dissolved hydrogen detection device can acquire dissolved hydrogen data from the water vapor in the four boiler tubes and transmit it to a processing device. The processing device can acquire and store the dissolved hydrogen data, power plant operating parameters, and boiler parameters. The processing device then filters and preprocesses the fault characteristic parameters in the power plant operating data to determine the optimal neural network prediction model. A corresponding tube wall temperature prediction model for the four boiler tubes is established based on the boiler parameters. An oxide film thickness calculation model is then established based on the tube wall temperature. Finally, a corrosion monitoring model is established based on the oxide film thickness, tube wall temperature, and dissolved hydrogen data. Finally, a fusion model is constructed by combining the optimal neural network prediction model with the corrosion monitoring model to obtain a fusion value. This fusion value is used to determine the leakage status of the four boiler tubes, thereby providing leak warnings with higher accuracy and reliability than existing technologies.

[0047] Optionally, the boiler four-tube leakage early warning system further includes a cockpit, which is communicatively connected to the processing equipment.

[0048] By adopting the above technical solution, the cockpit can display various data obtained by the processing equipment.

[0049] In summary, the present invention has at least the following beneficial technical effects:

[0050] The chemical water samples collected by this four-tube boiler leak warning method fully represent the operating conditions of the boiler's water-wall tubes, superheater tubes, reheater tubes, and economizer tubes. Because metal walls are constantly exposed to high-temperature water and steam, they produce trace amounts of dissolved hydrogen. Trace dissolved hydrogen is a corrosion product, mostly dissolved in water vapor, and can more accurately reflect the corrosion situation. A corrosion monitoring model, combined with long-term monitoring data of trace dissolved hydrogen, can effectively monitor and warn of four-tube boiler leaks. Furthermore, this four-tube boiler leak warning method does not require a full boiler deployment. By detecting trace dissolved hydrogen in samples of the boiler's water-wall tubes, superheater tubes, reheater tubes, and economizer tubes, it can effectively identify the oxidation status of the boiler tubes, significantly improving leak detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is the application architecture diagram of the boiler four-tube leakage early warning method;

[0052] Figure 2 This is the data flow diagram of the optimal neural network prediction model and corrosion monitoring model in the boiler four-tube leakage early warning method;

[0053] Figure 3 This is a schematic diagram of the structure of the optimal neural network prediction model in the boiler four-tube leakage early warning method;

[0054] Figure 4This is a schematic diagram of the structure of the tube wall temperature prediction model in the boiler four-tube leakage early warning method;

[0055] Figure 5 This is the structural diagram of the corrosion monitoring model in the boiler four-tube leakage early warning method;

[0056] Figure 6 This is a fusion model diagram of the optimal neural network prediction model and the corrosion model in the boiler four-tube leakage early warning method;

[0057] Figure 7 This is a graph showing the accuracy and loss changes of different network structures in Example 1;

[0058] Figure 8 This is a diagram showing changes in the steam temperature of the water wall outlet header and the induced draft fan opening degree caused by leakage in Example 1;

[0059] Figure 9 This is a comparison chart of the calculation results of the oxide film in Example 1;

[0060] Figure 10 This is a diagram showing the prediction results of key parameters in pipeline leakage in Example 1.

[0061] Explanation of the accompanying symbols: 1. dissolved hydrogen detection device; 2. processing equipment; 3. cockpit. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following Figure 1-10 It should be understood that the specific embodiments described herein are only used to illustrate the present invention and are not intended to limit the present invention.

[0063] An embodiment of the present invention provides a boiler four-tube leakage early warning method.

[0064] refer to Figure 1 , the boiler four-tube leakage early warning method includes the following steps:

[0065] S1: Obtain power plant operating parameters, and obtain dissolved hydrogen data of the four boiler tubes based on the dissolved hydrogen detection device 1.

[0066] During the operation of a thermal power plant, corresponding power plant operating parameters are generated, which can be obtained through a detection device installed on the thermal power plant. This is a prior art and will not be described in detail here. The power plant operating parameters are then obtained.

[0067] Dissolved hydrogen detection devices 1 are connected to each of the four boiler tubes. They detect dissolved hydrogen in the water vapor of these four boiler tubes, thereby acquiring corresponding dissolved hydrogen data. This data is then collected through multiple dissolved hydrogen detection devices 1. Specifically, dissolved hydrogen detection devices 1 are used to detect dissolved hydrogen data generated by the tubes, water-wall tubes, superheater tubes, and reheater tubes.

[0068] S2: Build Figure 2 The database middleware shown in the figure stores the power plant operating parameters and dissolved hydrogen data in the time series database through the data middleware.

[0069] Specifically, step S2 includes the following steps S21 and S22:

[0070] S21: The time series database uses the Timescale time series database, which is developed in Python and installs the ODBC (Open Database Connectivity) library to access the PI (Plant Information System) database and time series database to read, store, create, and delete databases and data.

[0071] S22: Encapsulates the database access API (Application Programming Interface), and the processor builds database middleware using Python, which includes an HTML5 model. Power plant operating parameters and dissolved hydrogen data are stored in a time-series database through the data middleware, enabling bidirectional real-time and historical data exchange between the simulation model and the physical equipment real-time database.

[0072] S3: Select fault characteristic parameters related to boiler four-tube leakage from the power plant operation parameters, and perform data preprocessing on the fault characteristic parameters.

[0073] Specifically, step S3 includes the following steps S31 to S34:

[0074] S31: Select fault characteristic parameters related to boiler four-tube leakage from the power plant operation parameters.

[0075] Fault characteristic parameters include: actual load, main steam pressure, feed water flow, induced draft impeller opening, left high-temperature superheater outlet steam temperature, right high-temperature superheater outlet steam temperature, left high-temperature superheater outlet steam pressure, right high-temperature superheater outlet steam pressure, left superheater secondary desuperheating water flow, right superheater secondary desuperheating water flow, left high-temperature reheater outlet steam temperature, right high-temperature reheater outlet steam temperature, left high-temperature reheater outlet steam pressure, right high-temperature reheater outlet steam pressure, left high-temperature reheater desuperheating water flow, right reheater desuperheating water flow, left economizer outlet flue gas temperature difference, economizer outlet feed water temperature, water-cooled wall outlet header temperature, and steam-water separator outlet steam pressure.

[0076] S32: Perform data cleaning on the fault characteristic parameters to obtain a cleaned data set.

[0077] The cleaned dataset can be obtained by preprocessing noisy, incomplete, and inconsistent fault feature parameters. Specifically, the preprocessing methods include missing value processing, erroneous data processing, and redundant data processing.

[0078] S33: Perform data dimensionality reduction on the cleaned data set to obtain low-dimensional feature parameters.

[0079] Principal Component Analysis (PCA) was used to zero-mean the cleaned data set. The covariance matrix was then calculated and the eigenvectors and eigenvalues ​​were derived. These eigenvectors and eigenvalues ​​form a new feature space, generating low-dimensional feature parameters. PCA effectively reduces the dimensionality of the existing fault characteristic parameters without losing important information, achieving data dimensionality reduction.

[0080] S34: Divide the low-dimensional feature parameters into training set, validation set and test set.

[0081] The low-dimensional feature parameters are normalized and divided into training, validation, and test sets. The training, validation, and test sets are stored in the Timescale time series database for easy access through the ODBC library.

[0082] S4: Determine the optimal neural network prediction model based on the fault characteristic parameters after data preprocessing.

[0083] Specifically, step S4 includes the following steps S41 and S42:

[0084] S41: Construct a CNN-LSTM neural network model (Convolutional Neural Network - Long Short-Term Memory) based on the training set, validation set, convolutional neural network model (CNN, Convolutional Neural Networks) and long short-term memory network model (LSTM, Long Short-Term Memory).

[0085] refer to Figure 3 First, we optimized the hyperparameters using the training set to ensure optimal performance of the CNN-LSTM neural network model. During this process, we carefully selected the appropriate optimizer (Adam), activation function (ReLU), and dropout rate to mitigate overfitting and improve the generalization capability of the CNN-LSTM neural network model.

[0086] Secondly, the process of training the CNN-LSTM neural network model includes several key steps: first, forward propagation, which passes the input data through the network and generates predictions; then loss calculation, which uses the loss function (using cross-entropy loss) to evaluate the gap between the output of the CNN-LSTM neural network model and the true label; then backpropagation, which uses the gradient descent method to update the weights in the CNN-LSTM neural network model; finally, iterative training, which repeats the above steps to gradually improve the performance of the CNN-LSTM neural network model.

[0087] Finally, monitor the training performance of the CNN-LSTM neural network model by observing the changes in the training set loss and validation set loss. If the changes in the training set loss and validation set loss remain within a small range and do not decrease significantly, the CNN-LSTM neural network model is close to convergence. Furthermore, if the accuracy on the validation set remains stable over multiple training rounds and does not increase significantly, this further indicates that the model is converging. Furthermore, if the validation set loss does not decrease after several iterations, training can be stopped to prevent overfitting.

[0088] S42: Determine the structural parameters of the optimal neural network prediction model based on the validation set, test set and CNN-LSTM neural network model.

[0089] We designed various 1D-CNN-LSTM neural network models, including 3C1L, 4C1L, and 4C2L architectures. The 1D-CNN-LSTM (3C1L) consists of three convolutional layers and one LSTM layer. The 1D-CNN-LSTM (4C1L) adds one convolutional layer and one pooling layer, while maintaining the same convolution kernel size. Furthermore, the 1D-CNN-LSTM (4C2L) builds on the 4C1L architecture by adding an LSTM layer.

[0090] Through multiple rounds of experiments, the training performance of different CNN-LSTM neural network models was compared, and the accuracy and loss values ​​of each CNN-LSTM neural network model after hyperparameter adjustment were recorded. Finally, the performance of the CNN-LSTM neural network model was evaluated on the validation set and test set, focusing on monitoring the generalization ability of the CNN-LSTM neural network model. From this, the hierarchical structure with stable performance, high accuracy, and low loss values ​​in the training and test sets was selected. Ultimately, the structural parameters of the CNN-LSTM neural network model with the best training effect were determined, that is, the optimal neural network prediction model was determined, and finally the h5 file of the optimal neural network prediction model was output.

[0091] S5: Obtain boiler parameters, and establish a tube wall temperature prediction model for the four tubes of the boiler based on the boiler parameters.

[0092] Specifically, step S5 includes the following steps S51 and S52:

[0093] S51: Obtain boiler parameters, and establish a combustion numerical simulation model based on the boiler parameters.

[0094] Boiler parameters include fuel characteristics, fuel flow, furnace dimensions, furnace temperature, pressure, and flue gas output. An appropriate meshing method was used to mesh these parameters. Considering that modeling the platen superheater and high-temperature superheater based on actual conditions could result in a significant increase in the number of meshes, an unstructured mesh was ultimately adopted after comprehensive consideration. The boundary conditions for the combustion numerical simulation model were then set. These included the primary air volume and temperature, the volume and temperature around the primary air nozzle, the exhaust gas volume and temperature, the volume and temperature around the exhaust gas nozzle, the secondary air volume distribution and temperature, and the burnout air volume and temperature. Furthermore, the turbulence intensity and hydraulic diameter were set for each inlet during the calculation process. For the primary air and exhaust gas, the combustion volume was also specified. Different wall boundary conditions were set for each region, and the outlet boundary condition was set as a pressure outlet boundary with a gauge pressure of -100 Pa. Finally, to ensure the reliability of the combustion numerical simulation model, the numerical simulation results were verified and compared.

[0095] Specifically, the combustion numerical simulation model is verified by comparing the simulation values ​​and design values ​​of the average flue gas temperatures at the inlet and outlet of boiler water-cooled wall tubes, superheater tubes, reheater tubes, and economizer tubes under various operating conditions such as BMCR (boiler maximum output condition), BRL (boiler rated load condition), THA (heat rate acceptance condition), 40%THA, 50%THA, and 75%THA.

[0096] S52: Based on the combustion numerical simulation model, Figure 4 The wall temperature prediction model of the four boiler tubes is shown.

[0097] By calibrating the simulation results using the design values ​​of the average inlet and outlet flue gas temperatures of the boiler water-wall tubes, superheater tubes, reheater tubes, and economizer tubes under various operating conditions, the wall temperature distribution of these tubes can be determined. A wall temperature prediction model for these four boiler tubes was then established based on a neural network algorithm. This algorithm eliminates the need to consider the angular coefficient when calculating wall temperature from flue gas temperature, thus reducing iterations and speeding up the calculation. This approach also unifies the wall temperature prediction models for the four heat exchangers.

[0098] Specifically, the temperature distribution data of boiler water-cooled wall tubes, superheater tubes, reheater tubes and economizer tube walls obtained under different operating conditions in the combustion numerical simulation model are used as training samples, and a neural network tube wall temperature prediction model for boiler water-cooled wall tubes, superheater tubes, reheater tubes and economizer is established. The test sample temperature data is compared with the predicted value to verify the accuracy of the tube wall temperature prediction model.

[0099] S6: Based on the tube wall temperature prediction model, establish an oxide film thickness calculation model. Based on the oxide film thickness calculation model, the tube wall temperature prediction model and the dissolved hydrogen data, establish Figure 5 The corrosion monitoring model shown.

[0100] Specifically, step S6 includes the following steps S61 to S63:

[0101] S61: Determine the outer wall temperatures of the four boiler tubes based on the tube wall temperature prediction model and the power plant operation parameters.

[0102] The tube wall temperature prediction model reads the on-site data of the thermal power plant operation through the data middleware, inputs the on-site data of the power plant operation such as boiler load, fuel consumption, primary air volume, primary air temperature, secondary air volume, secondary air temperature and the air volume percentage of each air layer, and calculates the outer wall temperature of the water-cooled wall tubes, superheater tubes, reheater tubes and economizer tubes.

[0103] S62: Establish an oxide film thickness calculation model based on the outer wall temperature of the four boiler tubes.

[0104] When the oxide film inside the four-tube boiler is evenly distributed, heat is transferred only in the radial direction, and is tightly bonded to the tube wall with no gaps in the middle, the oxide film thickness σ when the outer wall temperature is T is calculated using the tube wall temperature prediction model.

[0105] S63: Based on the oxide film thickness calculation model, pipe wall temperature prediction model and dissolved hydrogen data, the following is established: Figure 5 The corrosion monitoring model shown.

[0106] A corrosion monitoring model can be established based on the changing relationship between oxide film thickness and outer pipe temperature, as well as the online trace dissolved hydrogen content. The oxide film thickness is obtained from the oxide film thickness calculation model, the outer pipe temperature is obtained from the pipe wall temperature prediction model, and the dissolved hydrogen data is obtained from the online trace dissolved hydrogen by the dissolved hydrogen detection device 1.

[0107] Since the oxide film thickness changes significantly only every 100 hours, the calculated results of the outer wall temperature need to be input into the database, and the generated oxide film thickness is calculated every 24 hours. The corrosion monitoring model regularly monitors the water-cooled wall tubes, superheater tubes, reheater tubes, and economizer tubes for over-temperature warnings and determines whether the tube wall oxidation rate is accelerated based on the dissolved hydrogen content.

[0108] S7: Based on the optimal neural network prediction model and corrosion monitoring model construction Figure 6 The fusion model shown in FIG1 is based on a fusion model, which performs weighted averaging of redundant information from multiple sources to obtain a fusion value, and performs a boiler four-tube leakage warning based on the fusion value.

[0109] Using weighted data fusion technology, the results of the optimal neural network prediction model and the corrosion monitoring model are input to construct a fusion model. Based on the fusion model, the weighted average of multi-source redundant information is performed to obtain a fusion value, and the obtained fusion value is used to issue a boiler four-tube leakage warning.

[0110] In an optional embodiment of the present application, the boiler four-tube leakage early warning method further includes:

[0111] S8: Construct the cockpit 3 of the boiler four-tube leakage warning data and display the boiler four-tube leakage warning results.

[0112] Construct a cockpit 3 for boiler four-tube leakage warning data, output the leakage warning results of the optimal neural network prediction model, corrosion monitoring model and fusion model to display the alarm information in real time.

[0113] Specifically, the program of cockpit 3 includes an offline training module, a data storage module, an online prediction module, a data display module and an early warning module. Each module runs independently of each other and is called through a function.

[0114] The function call specifically involves calling the handle and request set within the program in the data storage module to pass data filtering criteria and obtaining the return data from the Timescale time series database using the getData function. In the online prediction module, real-time data is used as input, and prediction results are generated using the plotModel function. Finally, the online prediction module pushes the prediction results to the data display module for front-end display. Meanwhile, the early warning module uses the getAlarm function to obtain the alarm information (alarmInfo) output by the fusion model and displays it on the interface, allowing users to keep abreast of system status.

[0115] The online prediction module controls whether the online prediction model needs to be retrained offline through parameters. After receiving the retraining instruction output by the online prediction module, the offline training module retrains the CNN-LSTM neural network model and outputs the h5 algorithm file of the optimal neural network prediction model for use by the online prediction module.

[0116] Compared to existing technologies, this four-tube boiler leak warning method utilizes an electrochemical online dissolved hydrogen detection device 1, effectively avoiding the measurement point layout issues associated with acoustic emission devices and radiographic devices. In actual use, the dissolved hydrogen detection device 1 can be installed on the steam-water sampling rack in the water treatment workshop as needed. The feedwater, boiler water, or steam-water separator samples, main steam, and reheat steam samples corresponding to the boiler's water-wall tubes, superheater tubes, reheater tubes, and economizer tubes can then be connected to the dissolved hydrogen detection device 1. The water treatment workshop also has ample space for the dissolved hydrogen detection device 1.

[0117] The chemical water samples collected by this boiler four-tube leakage early warning method can fully represent the operating conditions of the boiler water-cooled wall tubes, superheater tubes, reheater tubes and economizer tubes. Since the metal wall surface is always accompanied by extremely trace corrosion in high-temperature water and steam, and extremely trace dissolved hydrogen is produced, trace dissolved hydrogen is a corrosion product, most of which is dissolved in water vapor. Dissolved hydrogen can more truly reflect the corrosion situation. Therefore, this boiler four-tube leakage early warning method monitors the power plant operating parameters for a long time through a fusion model established based on the changing relationship between the oxide film thickness and the tube wall temperature and the trace dissolved hydrogen content, which can effectively monitor and warn of oxidation leakage of the boiler four tubes.

[0118] Furthermore, this four-tube boiler leak warning method does not require full boiler deployment and can fully utilize dissolved hydrogen data from samples in the water treatment workshop for fault prediction. Furthermore, the corrosion monitoring model is only related to dissolved hydrogen, and the combustion conditions during boiler operation do not directly affect leak detection. Furthermore, by connecting data to the control room, this four-tube boiler leak warning method can provide real-time, dynamic data warnings for boiler leaks. This four-tube boiler leak warning method does not require any modifications to the heating surface, and historical data received from the water treatment workshop can be directly uploaded to the data center. Furthermore, the operating parameters of the optimal neural network prediction model are adaptable to the vast majority of supercritical boilers, making this four-tube boiler leak warning method sufficiently adaptable and applicable to different power plant boilers.

[0119] It should be noted that GB / T10184 "Technical Requirements for Industrial Boilers" and GB / T10185 "Performance Test Methods for Industrial Boilers" contain all the operating parameters that this boiler four-tube leakage warning method relies on, and can be found in all supercritical units of coal-fired power plants.

[0120] The present method is described below by way of Example 1:

[0121] In Example 1, the effectiveness of the boiler four-tube leakage early warning method is verified by combining the operating data of a domestic 600MW supercritical thermal power unit. The specific implementation steps are as follows:

[0122] First, according to step S1, power plant operating parameters and dissolved hydrogen data are obtained. Then, according to step S2, data middleware is built. Data cleaning is performed on the power plant operating parameters obtained in step S2. Noisy, incomplete, and inconsistent power plant operating parameters are preprocessed, including missing values, erroneous data, and redundant / correlated data. Principal component analysis (PCA) is then used to zero-mean the preprocessed power plant operating parameters. The covariance matrix is ​​then calculated, and the eigenvectors and eigenvalues ​​of the covariance matrix are then calculated. These eigenvectors form a new feature space. This reduces the dimensionality of the existing fault characteristic parameters without losing important information, thereby achieving data dimensionality reduction. Finally, historical data (with labels) from relevant operations is prepared, a data normalization method is determined, and the data is divided into training, validation, and test sets.

[0123] These include the following:

[0124] Data acquisition: The high level of automation in coal-fired power plants collects and stores a large amount of production process data and equipment operation data. The data exchange middleware directly reads the power plant database data, including operation data and chemical water parameters (this project uses the PI database system, which supports ODBC access).

[0125] Data storage: ODBC technology associated with the time series database (Timescale) is used to achieve two-way real-time and historical data interaction between the simulation model and the physical device real-time database.

[0126] Middleware Services: After installing the ODBC library, Python can access the PI database and the system's time series library, reading, storing, creating, and deleting database tables and data. Middleware services developed in Python enable users to interact with data.

[0127] Secondly, according to step S4, a suitable CNN-LSTM neural network model (DNN, DCNN, LSTM) is selected. Different CNN-LSTM neural network models have different prediction effects. Hyperparameters are determined for different CNN-LSTM neural network models.

[0128] Finally, the batch size and number of training times of each CNN-LSTM neural network model are determined according to step S4. The mean absolute error (MAE), root mean square error (RMSE) and score are used to evaluate each CNN-LSTM neural network model and generate the optimal neural network prediction model to complete the offline training.

[0129] In the case of the same data set, the performance of three different levels of 1D-CNN-LSTM, as well as DNN, CNN and LSTM neural network models were compared. 76% of the continuous time series data in the data set was used as the training set, and the remaining data was used as the test set. To improve training efficiency, the initial value of the learning rate was set to 0.005 during training, and the Adam algorithm was used to adjust the learning rate after each iteration. This method significantly suppressed the late volatility, allowing the CNN-LSTM neural network model to converge to the optimal solution faster. Figure 7 As shown in Figure 1, the training process of each CNN-LSTM neural network model in the test set is demonstrated.

[0130] Compared to DNN, CNN, and LSTM neural network models, the 1D-CNN-LSTM model outperforms in terms of training accuracy and loss. DNNs, unlike other neural network models, often struggle to converge during training due to their inability to effectively extract temporal features of the data, resulting in higher loss. Furthermore, the 1D-CNN and LSTM neural network models are relatively sensitive to local spatial and temporal features, and even with sufficient training cycles, their accuracy and loss still fluctuate to some extent. The 1D-CNN-LSTM neural network model, however, combines the advantages of CNN and LSTM to effectively capture the global and temporal features of the data while maintaining neural network performance, resulting in better prediction results. The final experimental results demonstrate that the 1D-CNN-LSTM neural network model significantly outperforms single CNN or LSTM neural network models in this task, demonstrating its effectiveness and superiority, and confirming it as the optimal neural network prediction model.

[0131] According to the leakage records of the power plant, the operator discovered a leak one day and shut down the plant for maintenance. This leak caused the shutdown and maintenance to last for 160 hours. The fault characteristic parameters during this operation period are as follows: Figure 8 As shown in Figure 2, when the normalized actual power remains at around 0.816, the normalized water wall outlet header steam temperature in the two sets of fault characteristic parameters remain at around 0.978 and 0.982, respectively, and the normalized induced draft fan opening remains at around 0.427 and 0.418, respectively. Figure 8 It can be seen that within the time period, the steam temperature of the water wall outlet header dropped to about 0.892 two days before the shutdown, and the induced draft fan blade opening increased to 0.482. At the same time, the fault characteristic parameters during this period are divided into three consecutive data sets, of which SetA and SetB are before the leakage, and SetC is after the leakage. At this time, the data sets of the corresponding time period are imported into the optimal neural network prediction model. The leakage detection results output by the optimal neural network prediction model are as follows: Figure 10 shown. Figure 10 In Figure c, we can see that the ms / mf ratio also increases. This phenomenon indicates that more feedwater is required for a certain steam flow rate. These phenomena are consistent with the behavior of a water wall leak.

[0132] Furthermore, due to fluctuations in the operating load of supercritical boilers, the false alarm rate of neural network leak detection is high. This paper establishes an adaptive alarm threshold for boiler four-tube leak detection by normalizing the relationship between power and mass flow, ensuring the accuracy of leak detection and minimizing the false alarm rate. Figure 10The figure illustrates the changes in key leak detection parameters after applying the adaptive alarm threshold method. The left subfigure shows the comparison between the predicted and measured values ​​of the key parameters after the adaptive alarm threshold is implemented. The right subfigure counts the cumulative number of times the predicted and measured values ​​exceed the adaptive threshold. In the normal operating data set, the percentage of predicted values ​​crossing the fixed threshold is 18% and 14%, respectively, while the percentage of measured values ​​is 24% and 21%. In the leaking operating data, the predicted and measured values ​​exceed the fixed threshold more frequently, at 42% and 34%, respectively. Under normal operating conditions, the percentage of predicted values ​​crossing the variable threshold decreases to 4% and 3%, respectively, while the percentage of measured values ​​crossing the variable threshold is 23% and 19%, respectively. Under fault conditions, the measured and predicted values ​​of the key parameters cross the variable threshold more frequently, at 37% and 31%, respectively. Therefore, the introduction of the adaptive alarm threshold significantly reduces the frequency of false alarms when the key parameters transition from one stable state to another. The variable threshold method provides a minimal false alarm rate under normal conditions while maintaining leak detection capability during fault conditions.

[0133] After the water-cooled wall leak occurred, the site was shut down for maintenance and a pipeline leak was discovered during the maintenance process. The crack at the water-cooled wall leak was small, about 56mm long and about 7mm wide at its widest point. The fracture surface of the crack was rough and uneven, with blunt edges and a large number of longitudinal cracks around it. Analysis showed that the leak was caused by high-temperature creep of the material after the pipeline was overheated for a long time. Pipeline leakage accidents are usually sudden processes. The optimal neural network prediction model can detect abnormal signs in advance, thereby reducing the losses caused by leakage. Figure 10 As shown in Figure 3, this case illustrates that fault characteristic parameters can be used to detect abnormalities in the on-site operating data of a coal-fired power plant in advance.

[0134] According to step S6, a corrosion monitoring model is established, and a numerical combustion simulation of the W-type flame boiler furnace is completed through Fluent. The simulation results are verified by the design values ​​of the average inlet and outlet flue gas temperatures of the boiler water-cooled wall, superheater, reheater, and economizer under various operating conditions, thereby obtaining the wall temperature distribution of the boiler water-cooled wall tubes, superheater tubes, reheater tubes, and economizer tubes under various operating conditions. Then, a neural network algorithm is used to establish a tube wall temperature prediction model for the four boiler tubes.

[0135] First, a CFD furnace combustion numerical simulation model was designed based on boiler parameters. Taking the high-temperature superheater as an example, the model included exhaust steam tuyere, main primary air tuyere, oil gun tuyere, three layers of secondary air distribution on the front and rear walls under the arch, burnout air nozzles above the main combustion zone, a screen superheater located in the upper furnace, and a high-temperature superheater within the horizontal flue. After setting an appropriate number of grid cells, the inlet, avoidance, and outlet boundary conditions were determined. The CFD calculation results were verified by comparing the average inlet and outlet temperatures of the high-temperature superheater with the design values.

[0136] Table 1 Comparison of errors in CFD calculation results for main working conditions

[0137]

[0138] Secondly, the wall temperature distribution data for the boiler water-wall tubes, superheater tubes, reheater tubes, and economizer tubes, obtained under different operating conditions in the simulation model, is stored in the system via the aforementioned middleware service. Data such as boiler load, fuel consumption, primary air volume and temperature, secondary air volume and temperature, and the air volume percentages for each air layer are input into a deep neural network model, which outputs predicted wall temperature distribution values. The neural network wall temperature distribution prediction model for the boiler water-wall tubes, superheater tubes, reheater tubes, and economizer tubes is trained on a test dataset to improve prediction accuracy. Compared to traditional wall temperature calculations based on flue gas heat load calculation models and working fluid measurement mechanism models, obtaining wall temperature distribution predictions based on neural network training reduces complex operations such as iterative operations and pipe section angle coefficient calculations, thereby improving prediction speed. Based on the piping material and creep temperature data in the high-temperature superheater, the two temperatures can be directly compared, providing immediate overtemperature warnings.

[0139] Finally, the oxide film temperature and growth rate calculations were incorporated into the wall temperature calculation program to obtain the oxide film thickness calculation results. After coupling the wall temperature calculation results, the oxide film thickness calculation was coupled with the wall temperature calculation. The higher the inner wall temperature of the boiler water-cooled wall tubes, superheater tubes, reheater tubes, and economizer tubes, the faster the oxide film forms. As the oxide film thickness increases, the outer wall temperature also increases, and even overtemperature creep occurs. Furthermore, the metal matrix on the steam side of the heating surface corrodes and forms an oxide film, causing the tube wall thickness to continuously decrease. Ultimately, when the stress on the tube wall exceeds the critical stress, the oxide film will flake off. The flakes of oxide scale can easily clog the pipe elbows, causing the elbow wall temperature to rise and causing overtemperature tube bursts.

[0140] Under the assumption that the oxide film inside the tube is uniformly distributed, heat transfers only in the radial direction, axial heat transfer is not considered, and it is tightly bonded to the metal tube wall with no gaps in the middle, the wall temperature calculation program calculates the oxidation temperature T when the oxide film thickness is σ. The oxide film thickness σ is calculated based on the oxidation temperature T and the oxidation time Δτ, and the superheater operation time τ is updated, and finally the operation reaches the set time. The results of the oxide film calculation model are obtained when the working fluid temperature is 600℃. Compared with the literature data, it can be seen that Figure 9 As shown in the figure, the calculated results are close to the experimental data in the literature. The oxide film calculation model can obtain relatively accurate data and be applied in engineering projects.

[0141] The fusion model combines the results of the optimal neural network prediction model and the corrosion monitoring model and outputs the prediction results, forming a "four-pipe" leakage and early warning system fusion model based on real-time monitoring and deep learning through big data analysis. Weighted averaging methods include weighted averaging, Kalman filtering, Bayesian estimation, fuzzy logic, neural networks, etc. This project uses the weighted averaging method for information fusion fault prediction. Ultimately, the system's software program is divided into an offline training module, a data storage module, an online prediction module, and a display / drawing module. Each module operates independently and is called through functions. The offline training module saves the output evaluation parameters of the trained model as an h5 file and provides it to the online prediction module. The online prediction module uses parameters to control whether the online prediction model needs to be retrained offline.

[0142] An embodiment of the present invention also provides a boiler four-tube leakage early warning system.

[0143] The four-tube boiler leak warning system includes a dissolved hydrogen detection device 1 and processing equipment 2. The dissolved hydrogen detection device 1 is connected to the economizer tubes, water-wall tubes, superheater tubes, and reheater tubes, and is in communication with the processing equipment 2. The dissolved hydrogen detection device 1 is used to detect dissolved hydrogen generated by the economizer tubes, water-wall tubes, superheater tubes, and reheater tubes. The processing equipment 2, as the executor of this four-tube boiler leak warning method, is responsible for performing all steps except detecting dissolved hydrogen and displaying various data in the cockpit 3. The processing equipment 2 can be implemented in various forms, including mobile phones, tablet computers, PDAs, laptop computers, and desktop computers.

[0144] The boiler four-tube leakage warning system further includes a cockpit 3, which is in communication with the processing device 2. It should be understood that the cockpit 3 can be a display device such as a display screen.

[0145] The various variations and specific examples of the boiler four-tube leakage warning method provided in the above embodiment are applicable to the two boiler four-tube leakage warning systems in this embodiment. Through the above detailed description of the boiler four-tube leakage warning method, those skilled in the art can clearly understand the implementation method of the boiler four-tube leakage warning system in this embodiment. For the sake of brevity of the specification, it will not be described in detail here.

[0146] As described above, the above embodiments are only used to provide a detailed introduction to the technical solutions of the present invention. However, the description of the above embodiments is only used to help understand the method and core concept of the present invention and should not be understood as limiting the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention.

Claims

1. A boiler four-tube leakage early warning method, characterized in that: The following steps are involved: S1: Obtaining power plant operating parameters and obtaining dissolved hydrogen data of the four boiler tubes based on the dissolved hydrogen detection device (1); S2: Build database middleware and store power plant operating parameters and dissolved hydrogen data in a time series database through the data middleware; S3: Select fault characteristic parameters related to boiler four-tube leakage from the power plant operation parameters, and perform data preprocessing on the fault characteristic parameters; S4: Determine the optimal neural network prediction model based on the fault characteristic parameters after data preprocessing; The step S4 comprises the following steps: S41: Construct a CNN-LSTM neural network model based on the training set, validation set, convolutional neural network model, and long short-term memory network model; S42: Determine the optimal neural network prediction model based on the validation set, test set and CNN-LSTM neural network model; S5: Obtain boiler parameters and establish a wall temperature prediction model for the four boiler tubes based on the boiler parameters; S6: Establish an oxide film thickness calculation model based on the pipe wall temperature prediction model, and establish a corrosion monitoring model based on the oxide film thickness calculation model, the pipe wall temperature prediction model, and the dissolved hydrogen data; S7: A fusion model is constructed based on the optimal neural network prediction model and the corrosion monitoring model. The multi-source redundant information is weighted averaged based on the fusion model to obtain a fusion value. The boiler four-tube leakage warning is performed based on the fusion value.

2. The boiler four-tube leakage early warning method according to claim 1, characterized in that: The step S3 comprises the following steps: S31: selecting fault characteristic parameters related to boiler four-tube leakage from power plant operation parameters; S32: performing data cleaning on the fault characteristic parameters to obtain a cleaned data set; S33: Perform data dimensionality reduction on the cleaned data set to obtain low-dimensional feature parameters; S34: Divide the low-dimensional feature parameters into training set, validation set and test set.

3. The boiler four-tube leakage early warning method according to claim 2, characterized in that: The fault characteristic parameters include: actual load, main steam pressure, feed water flow, induced draft impeller opening, left high-temperature superheater outlet steam temperature, right high-temperature superheater outlet steam temperature, left high-temperature superheater outlet steam pressure, right high-temperature superheater outlet steam pressure, left superheater secondary cooling water flow, right superheater secondary cooling water flow, left high-temperature reheater outlet steam temperature, right high-temperature reheater outlet steam temperature, left high-temperature reheater outlet steam pressure, right high-temperature reheater outlet steam pressure, left high-temperature reheater cooling water flow, right reheater cooling water flow, left and right economizer outlet flue gas temperature difference, economizer outlet feed water temperature, water-cooled wall outlet header temperature, and steam-water separator outlet steam pressure.

4. The boiler four-tube leakage early warning method according to claim 1, characterized in that: The step S5 comprises the following steps: S51: Obtain boiler parameters and establish a combustion numerical simulation model based on the boiler parameters; The boiler parameters include fuel characteristics, fuel flow, furnace size, furnace temperature, pressure, and flue gas output. The boiler parameters are divided using an unstructured grid. The boundary conditions of the combustion numerical simulation model include primary air volume and temperature, air volume and temperature around the primary air nozzle, exhaust gas volume and temperature, air volume and temperature around the exhaust gas nozzle, secondary air volume distribution and secondary air temperature, and burnout air volume and temperature. S52: Establish a wall temperature prediction model for the four boiler tubes based on the combustion numerical simulation model.

5. The boiler four-tube leakage early warning method according to claim 4, characterized in that: The combustion numerical simulation model is verified by comparing the simulation values ​​and design values ​​of the average flue gas temperatures at the inlet and outlet of the boiler water-cooled wall tubes, superheater tubes, reheater tubes, and economizer tubes under the boiler maximum output condition, boiler rated load condition, heat rate acceptance condition, 40%THA, 50%THA, and 75%THA conditions.

6. The boiler four-tube leakage early warning method according to claim 4, characterized in that: The step S6 comprises the following steps: S61: Determine the outer wall temperature of the four boiler tubes based on the tube wall temperature prediction model and the power plant operation parameters; S62: Establish an oxide film thickness calculation model based on the outer wall temperature of the four boiler tubes; S63: A corrosion monitoring model is established based on the oxide film thickness calculation model, the pipe wall temperature prediction model, and the dissolved hydrogen data.

7. The boiler four-tube leakage early warning method according to any one of claims 1 to 6, characterized in that: The following steps are also included: S8: Construct a cockpit (3) for boiler four-tube leakage warning data and display the boiler four-tube leakage warning results.

8. A boiler four-tube leakage early warning system, characterized in that: The invention comprises a dissolved hydrogen detection device (1) and a processing device (2), wherein the dissolved hydrogen detection device (1) is connected to the economizer tube, the water-cooled wall tube, the superheater tube and the reheater tube, and the processing device (2) is communicatively connected to the dissolved hydrogen detection device (1). The processing device (2) comprises a processor, a memory and a communication bus, wherein the communication bus is used to realize the communication connection between the processor and the memory, and the processor is used to execute a computer program stored in the memory to realize the boiler four-tube leakage early warning method according to any one of claims 1 to 7.

9. The boiler four-tube leakage warning system according to claim 8, characterized in that: It also includes a cockpit (3), which is communicatively connected to the processing device (2).

Images