Method and system for determining water power characteristics of boiler water wall based on CFD-BP coupling

Abstract

This invention belongs to the field of determining the hydrodynamic characteristics of water-cooled walls, and discloses a method and system for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling. This invention constructs a CFD numerical simulation model coupling the flue gas side of the furnace and the working fluid side inside the water-cooled wall tubes, generating a simulation sample set containing characteristic data of different operating conditions and hydrodynamic parameters of each water-cooled wall tube. This expands the operating conditions covered by hydrodynamic characteristic analysis and provides training data with clear physical correspondences for the prediction model. By training a BP neural network with the simulation sample set, a hydrodynamic characteristic prediction model is obtained. This allows for the rapid acquisition of hydrodynamic parameters of each water-cooled wall tube without repeatedly performing complex CFD solutions for each tested operating condition, thus balancing the computational accuracy and processing efficiency of hydrodynamic characteristic determination.

Description

Technical Field

[0001] This invention belongs to the field of determining the hydrodynamic characteristics of water-cooled walls, specifically involving a method and system for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling. Background Technology

[0002] With the construction of new power systems and the increasing demand for deep peak-shaving operation of coal-fired units, boiler load and coal quality conditions change frequently, leading to fluctuations in the spatial distribution of furnace heat load. As a crucial heat-bearing and pressure-bearing component of coal-fired boilers, the hydrodynamic characteristics of water-cooled walls—including the working fluid mass flow rate, pressure drop, cavitation fraction, outlet enthalpy, tube wall temperature, and flow stability—directly affect whether flow stagnation, backflow, localized overheating, or tube rupture occurs. Therefore, accurately determining the hydrodynamic characteristics of each water-cooled wall tube under different operating conditions is of great significance.

[0003] Existing methods for determining the hydrodynamic characteristics of water-cooled walls mainly include field measurements, one-dimensional pipe network analytical calculations, three-dimensional CFD numerical simulations, and BP neural network predictions based on operational data. Field measurements typically utilize differential pressure and flow data from the DCS system or pre-embedded thermocouples to obtain operating parameters. However, due to limitations in the number of measuring points, layout conditions, and high-temperature and high-pressure environments, only a small number of measuring points or loop average data can usually be obtained, making it difficult to cover all water-cooled wall tubes. Furthermore, the cost of modification and maintenance is relatively high.

[0004] One-dimensional pipe network analytical calculations are fast, but they often use simplified conditions such as average heat load, making it difficult to reflect the three-dimensional spatial distribution of heat transfer and heat load on the flue gas side of the furnace. They also lack adaptability to low-load, combustion skew, and two-phase flow nonlinear conditions. Three-dimensional CFD numerical simulations can solve the heat transfer and flow processes between the flue gas side of the furnace and the working fluid side of the water-cooled wall tubes with greater precision, but the multi-scale mesh and multi-physics coupling result in a large computational load, making it difficult to meet the requirements for batch calculations under multiple operating conditions and online real-time measurement.

[0005] Backpropagation neural networks trained on historical DCS operating data offer fast prediction speeds, but they suffer from limited extreme condition samples and lack constraints from physical laws such as mass and energy conservation, making them prone to prediction distortion. Existing methods combining CFD and neural networks typically rely solely on offline training of the network using CFD results. They lack a closed-loop mechanism for correcting numerical models and network parameters based on actual field measurement deviations, and the averaging of parameters across multiple output loops makes it difficult to determine the hydrodynamic parameters of each water-cooled wall tube individually. Summary of the Invention

[0006] The purpose of this invention is to overcome the problems of insufficient full-condition sample coverage, difficulty in balancing computational accuracy and real-time performance, easy decay of model prediction accuracy, and difficulty in accurately identifying the operating risks of individual water-cooled wall tubes in existing methods for determining the hydrodynamic characteristics of boiler water-cooled walls. This invention provides a method and system for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling, comprising the following steps: Obtain boiler structural parameters and operating parameters under different working conditions, construct a CFD numerical simulation model that couples the flue gas side of the furnace with the working fluid side inside the water-cooled wall tubes, and use the CFD numerical simulation model to generate a simulation sample set containing operating condition characteristic data and hydrodynamic parameters of each water-cooled wall tube. Using operating condition characteristic data as input and hydrodynamic parameters of each water-cooled wall tube as output, a BP neural network is trained to obtain a hydrodynamic characteristic prediction model. The working condition characteristic data of the test condition are obtained and input into the hydrodynamic characteristic prediction model to obtain the hydrodynamic parameters of each water-cooled wall tube; Based on hydrodynamic parameters and preset safety judgment conditions, identify water-cooled wall tubes with operational risks and output the risk judgment results; Obtain on-site measured data of the working condition to be tested, correct the CFD numerical simulation model based on the deviation between the on-site measured data and the corresponding hydrodynamic parameters, generate supplementary simulation samples using the corrected CFD numerical simulation model, and update the hydrodynamic characteristic prediction model.

[0008] A further improvement of this invention lies in the following method: The method for obtaining boiler structural parameters and operating parameters under different conditions, constructing a CFD numerical simulation model coupling the flue gas side of the furnace and the working fluid side inside the water-cooled wall tubes, and generating a simulation sample set containing operating condition characteristic data and hydrodynamic parameters of each water-cooled wall tube using the CFD numerical simulation model is as follows: Based on the boiler structural parameters, a computational domain for the flue gas side of the furnace and a computational domain for the working fluid side of each water-cooled wall tube are established. The computational domains for the flue gas side and the working fluid side are divided into multi-scale grids, and heat flow data is transferred through the interface. The flue gas flow and radiation heat transfer processes in the computational domain on the flue gas side of the furnace, as well as the working fluid flow, heat transfer, and phase change processes in the computational domain on the working fluid side, are solved in a coupled manner. Different operating conditions are set according to boiler load, feedwater parameters, fuel parameters, air volume parameters, coal quality parameters and combustion deviation parameters, and CFD numerical simulation models are used to solve the different operating conditions respectively. The operating condition characteristic data corresponding to each operating condition are extracted from the solution results, as well as the working fluid mass flow rate, pressure drop, cavitation fraction, outlet working fluid enthalpy, tube wall temperature, local heat load, critical heat flux density, flow instability judgment index, backflow critical flow rate and thermal deviation coefficient corresponding to each water-cooled wall tube, to form a simulation sample set.

[0009] A further improvement of this invention lies in the following method for training a BP neural network using operating condition characteristic data as input and hydrodynamic parameters of each water-cooled wall tube as output, to obtain a hydrodynamic characteristic prediction model: The working condition characteristic data and hydrodynamic parameters in the simulation sample set are normalized, and the normalized simulation sample set is divided into training set, validation set and test set. A multi-output BP neural network is constructed, consisting of an input layer, at least one hidden layer, and an output layer. The neurons in the input layer correspond to the operating condition feature data, and the neurons in the output layer correspond to the hydrodynamic parameters of each water-cooled wall tube. A composite loss function is constructed based on prediction error loss, mass conservation loss, and energy conservation loss. The multi-output BP neural network is trained using the training set and the composite loss function. The training termination condition is determined using the validation set, and the training results are verified using the test set. Save the network weights, bias parameters, and normalized mapping parameters of the trained multi-output BP neural network to obtain the hydrodynamic characteristic prediction model.

[0010] A further improvement of this invention lies in the following method for obtaining the operating condition characteristic data of the test condition and inputting it into the hydrodynamic characteristic prediction model to obtain the hydrodynamic parameters of each water-cooled wall tube: Acquire at least one of the following parameters for the operating condition under test: boiler load, feedwater parameters, fuel parameters, air volume parameters, coal quality parameters, furnace heat load parameters, and water-cooled wall inlet parameters, to form operating condition characteristic data for the operating condition under test. The working condition characteristic data of the test condition are normalized according to the normalized mapping parameters, and the normalized working condition characteristic data are input into the hydrodynamic characteristic prediction model for forward propagation. The prediction results output by the hydrodynamic characteristic prediction model are inversely normalized. The correspondence between the prediction results and each water-cooled wall tube is established according to the pipe number of the water-cooled wall tube, and the hydrodynamic parameters of each water-cooled wall tube are obtained.

[0011] A further improvement of this invention lies in the following method for identifying water-cooled wall tubes with operational risks based on hydrodynamic parameters and preset safety judgment conditions, and outputting the risk judgment results: The thermal deviation coefficient, flow instability judgment index, working fluid mass flow rate and maximum pipe wall temperature of each water-cooled wall tube are compared with the corresponding safety judgment threshold. When the thermal deviation coefficient of the water-cooled wall tube exceeds the thermal deviation threshold, it is determined that the water-cooled wall tube has a high risk of thermal deviation. When the flow instability determination index of the water-cooled wall tube exceeds the flow stability threshold, it is determined that the water-cooled wall tube has a risk of two-phase flow pulsation. When the working fluid mass flow rate of the water-cooled wall tube is lower than the corresponding critical backflow flow rate, it is determined that there is a risk of flow stagnation or backflow in the water-cooled wall tube. When the maximum wall temperature of the water-cooled wall tube exceeds the allowable temperature of the corresponding tube material, it is determined that there is a risk of local overheating in the water-cooled wall tube. Based on the comparison results, determine the risk type, risk level, and safety margin of each water-cooled wall tube, and output the risk assessment results including the dangerous pipe number, furnace location, risk type, and safety margin.

[0012] A further improvement of this invention lies in obtaining on-site measured data of the working condition to be tested, correcting the CFD numerical simulation model based on the deviation between the on-site measured data and the corresponding hydrodynamic parameters, and using the corrected CFD numerical simulation model to generate supplementary simulation samples to update the hydrodynamic characteristic prediction model. The specific method is as follows: The pipe wall temperature, total loop pressure difference, and total water supply flow rate under the same test condition and at the same acquisition time are used as field measured data. The hydrodynamic parameters of each water-cooled wall tube are screened or summarized to obtain predicted data corresponding to the field measured data, and the deviation between the predicted data and the field measured data is calculated. Correct at least one of the following model parameters in the CFD numerical simulation model based on the deviation: furnace radiation heat transfer coefficient, water-cooled wall two-phase flow boiling heat transfer coefficient, and wall heat flux scaling coefficient. The modified CFD numerical simulation model is used to solve the supplementary working conditions, generating supplementary simulation samples. These supplementary simulation samples are then added to the simulation sample set, and the updated simulation sample set is used to update the hydrodynamic characteristic prediction model.

[0013] Secondly, the present invention provides a system for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling, comprising: The simulation sample construction module is used to obtain boiler structural parameters and operating parameters under different working conditions, construct a CFD numerical simulation model that couples the flue gas side of the furnace with the working fluid side inside the water-cooled wall tubes, and use the CFD numerical simulation model to generate a simulation sample set containing operating condition characteristic data and hydrodynamic parameters of each water-cooled wall tube. The prediction model training module is used to train the BP neural network with the operating condition feature data as input and the hydrodynamic parameters of each water-cooled wall tube as output, so as to obtain the hydrodynamic characteristic prediction model. The hydrodynamic parameter determination module is used to acquire the working condition characteristic data of the working condition to be tested and input it into the hydrodynamic characteristic prediction model to obtain the hydrodynamic parameters of each water-cooled wall tube. The risk assessment module is used to identify water-cooled wall tubes with operational risks based on hydrodynamic parameters and preset safety assessment conditions, and output the risk assessment results. The feedback update module is used to acquire the field measured data of the working condition to be tested, correct the CFD numerical simulation model based on the deviation between the field measured data and the corresponding hydrodynamic parameters, generate supplementary simulation samples using the corrected CFD numerical simulation model, and update the hydrodynamic characteristic prediction model.

[0014] A further improvement of this invention is that the simulation sample construction module includes: The computational domain construction unit is used to establish the computational domain on the flue gas side of the furnace and the computational domain on the working fluid side of each water-cooled wall tube based on the boiler structural parameters, and to establish the heat flow data transfer relationship between the computational domain on the flue gas side of the furnace and the computational domain on the working fluid side. The batch solution unit is used to receive operating parameters under different working conditions, set the operating parameters as boundary conditions of the CFD numerical simulation model, and perform coupled solution for different working conditions. The sample extraction unit is used to extract operating condition feature data and hydrodynamic parameters of each water-cooled wall tube from the solution results output by the batch solution unit, form a simulation sample set, and transmit the simulation sample set to the prediction model training module.

[0015] A further improvement of the present invention is that the prediction model training module is also used to transmit the network weights, bias parameters and normalized mapping parameters of the hydrodynamic characteristic prediction model to the hydrodynamic parameter determination module. The hydrodynamic parameter determination module is connected to the boiler distributed control system to receive the operating condition characteristic data of the operating condition to be measured, process the operating condition characteristic data according to the normalized mapping parameters, and generate the hydrodynamic parameters of each water-cooled wall tube using network weights and bias parameters. The hydrodynamic parameter determination module is also used to transmit the hydrodynamic parameters of each water-cooled wall tube to the risk assessment module, so that the risk assessment module can generate a risk assessment result that includes the dangerous pipeline number, furnace location, risk type and safety margin.

[0016] A further improvement of the present invention is that the feedback update module is communicatively connected to the boiler distributed control system, the hydrodynamic parameter determination module, the simulation sample construction module, and the prediction model training module, respectively. The feedback update module is used to receive the field measured data output by the boiler distributed control system and the hydrodynamic parameters output by the hydrodynamic parameter determination module, determine the predicted data corresponding to the field measured data, and calculate the deviation between the predicted data and the field measured data. The feedback update module is also used to transmit model parameter correction data to the simulation sample construction module according to the deviation, so that the simulation sample construction module can generate supplementary simulation samples using the corrected CFD numerical simulation model, and transmit the supplementary simulation samples to the prediction model training module. The prediction model training module is used to update the hydrodynamic characteristic prediction model using supplementary simulation samples, and then transmits the updated hydrodynamic characteristic prediction model to the hydrodynamic parameter determination module.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs a CFD numerical simulation model coupling the flue gas side of the furnace and the working fluid side of the water-cooled wall tubes. It generates a simulation sample set containing characteristic data of different operating conditions and hydrodynamic parameters of each water-cooled wall tube. This expands the operating conditions covered by hydrodynamic characteristic analysis and provides training data with clear physical correspondences for the prediction model. By training a BP neural network with the simulation sample set, a hydrodynamic characteristic prediction model is obtained. This model can quickly obtain the hydrodynamic parameters of each water-cooled wall tube without repeatedly performing complex CFD solutions for each tested operating condition, thus balancing the computational accuracy and processing efficiency of hydrodynamic characteristic determination. By outputting the hydrodynamic parameters separately for each water-cooled wall tube and combining them with preset safety judgment conditions to identify water-cooled wall tubes with operational risks, refined identification of risk locations can be achieved, improving the targeting of water-cooled wall operation status monitoring and risk judgment. Furthermore, the CFD numerical simulation model is corrected based on the deviation between the on-site measured data and the corresponding hydrodynamic parameters. The corrected model is then used to generate supplementary simulation samples to update the hydrodynamic characteristic prediction model, forming a data closed loop of simulation, prediction, correction, and updating. This can reduce the deviation between the model calculation results and the actual operating state, improve the model's adaptability to changes in boiler operating state, and enhance the reliability of subsequent hydrodynamic characteristic determination results. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention; Figure 2 This is a system diagram of the present invention. Detailed Implementation

[0019] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0020] Example 1: This embodiment provides a method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling. The method generates a simulation sample set covering different operating conditions using a CFD numerical simulation model, trains a BP neural network using the simulation sample set, determines the hydrodynamic parameters of each water-cooled wall tube under the tested operating condition based on the trained hydrodynamic characteristic prediction model, assesses the operational risk of a single tube based on the hydrodynamic parameters, and updates the CFD numerical simulation model and the hydrodynamic characteristic prediction model using field measurement data.

[0021] See Figure 1This embodiment specifically includes the following steps: S101: Obtain boiler structural parameters and operating parameters under different working conditions. Based on the boiler structural parameters, establish the calculation domain of the flue gas side of the furnace and the calculation domain of the working fluid side of each water-cooled wall tube, and construct a CFD numerical simulation model that couples the flue gas side of the furnace and the working fluid side of the water-cooled wall tube.

[0022] A one-to-one geometric model is created based on the actual structure of the boiler furnace and water-cooled walls. The boiler structural parameters include furnace dimensions, water-cooled wall arrangement, number of water-cooled wall tubes, pipe number of each water-cooled wall tube, pipe length, pipe inner diameter, pipe wall thickness, loop connection relationship, and elbow position.

[0023] Establish a computational domain for the furnace flue gas side and a computational domain for the working fluid side corresponding to each water-cooled wall tube. The computational domain for the furnace flue gas side is used to calculate the flow of flue gas inside the furnace, the transport of combustion products, and the radiative heat transfer between the flue gas and the water-cooled wall; the computational domain for the working fluid side is used to calculate the flow, heat transfer, and phase change processes of the working fluid inside the water-cooled wall tube.

[0024] In one specific embodiment, the computational domain on the flue gas side of the furnace adopts a structured grid with a size of 0.2m to 0.5m, while the single-tube channel of the water-cooled wall adopts a boundary layer-refined grid with a size of 1mm to 3mm. By refining the grid in the near-wall region of the water-cooled wall tube, the changes in the working fluid velocity, temperature, and phase content in the near-wall region are characterized.

[0025] Since the computational domains on the flue gas side and the working fluid side have different grid scales, an interface is set between them. The heat flux data on the flue gas side and the wall temperature data on the working fluid side are transmitted through interpolation at the interface, so as to achieve data matching between computational domains of different scales and reduce the transmission error caused by the difference in grid scale.

[0026] The Realizable k-ε turbulence model was used to calculate the turbulent flow of flue gas on the furnace side, and the P1 radiation model was used to calculate the radiative heat transfer between flue gas, furnace wall and water-cooled wall.

[0027] The distribution of liquid and vapor phases and the phase interface changes inside the water-cooled wall tube are characterized by the VOF two-phase flow model. The working fluid properties under different pressure, temperature and enthalpy conditions are determined by calling the subcritical working fluid thermophysical property database or the supercritical working fluid thermophysical property database.

[0028] The CFD numerical simulation model solves the continuity equation, momentum equation, energy equation, and two-phase mass transfer conservation equation in a coupled manner. The wall heat flux calculated on the furnace flue gas side is transferred to the working fluid side of the water-cooled wall tubes through the interface, serving as the thermal boundary for calculating the heat absorption of the working fluid inside the tubes. The tube wall temperature calculated on the working fluid side is returned to the furnace flue gas side to update the radiative and convective heat transfer conditions, thereby achieving a coupled solution between the furnace flue gas side and the working fluid side of the water-cooled wall tubes.

[0029] In practical applications, CFD numerical simulation models can be built and solved using Fluent, or coupled numerical simulations of the furnace flue gas side and the working fluid side inside the water-cooled wall tubes can be completed using STAR-CCM+ or ANSYS CFX.

[0030] Different operating conditions are set based on boiler load, feedwater parameters, fuel parameters, air volume parameters, coal quality parameters, and combustion deviation parameters. These different operating conditions cover normal operation, deep peak shaving, combustion skew, and low flow limit conditions.

[0031] In one specific embodiment, the boiler load range is set to 20%–100% BMCR, the coal volatile matter range is set to 12%–32%, the feedwater pressure range is set to 16MPa–32MPa, and the primary air temperature range is set to 220℃–350℃. By changing the boiler load, coal quality, feedwater conditions, air distribution conditions, burner operating status, and four-corner combustion deviation, 1200 sets of differentiated operating conditions are generated in batches.

[0032] The 1200 sets of operating conditions simultaneously cover normal and stable operating states as well as boundary states that may cause flow stagnation, backflow, thermal deviation, or local overheating, thereby using CFD numerical simulation models to form extreme operating condition samples that are difficult to obtain from field operating data.

[0033] Operating condition sampling can employ either tiered sampling or uniform combination methods within a parameter range. In other embodiments, optimal Latin hypercube sampling or Monte Carlo constrained sampling can also be used to generate different combinations of operating conditions under given parameter ranges and constraints.

[0034] Different operating conditions were solved in parallel or sequentially using CFD numerical simulation models. Operating condition characteristic data were extracted from the solution results of each set of operating conditions, and the hydrodynamic parameters of each water-cooled wall tube were extracted according to its pipe number.

[0035] In one specific embodiment, the operating condition feature data includes 28-dimensional input features, specifically including: Boiler load, feedwater temperature, feedwater pressure, total fuel quantity, primary air volume, secondary air volume, pulverized coal injection volume at each layer, flue gas temperature at furnace outlet, average heat flux in furnace, total flow rate at water-cooled wall inlet, calorific value of coal, ash content of coal, excess air coefficient, burner tilt angle, recirculation air volume, desuperheating water volume, furnace negative pressure, initial tube wall temperature, working fluid saturation temperature, pipe length, pipe inner diameter, pipe wall thickness, flue gas side heat transfer coefficient, radiative heat flux ratio, working fluid inlet enthalpy, total pressure drop in the loop, and four-corner combustion deviation coefficient.

[0036] The pulverized coal injection rate for each combustion layer is expanded into corresponding input dimensions, making the above parameters constitute 28-dimensional operating condition characteristic data. The average heat flux of the furnace belongs to the furnace heat load parameter; the total flow rate at the water-cooled wall inlet, the enthalpy at the working fluid inlet, and the saturation temperature of the working fluid belong to the water-cooled wall inlet parameters.

[0037] The hydrodynamic parameters include 12 hydrodynamic indicators for each water-cooled wall tube, specifically: Single-pipe working fluid mass flow rate, frictional pressure drop along the pipe, local elbow pressure drop, average cavitation fraction of two-phase flow, enthalpy of working fluid at pipe outlet, average pipe wall temperature, maximum pipe wall temperature, local heat load peak value, critical heat flux density, flow instability judgment index, critical flow rate for backflow, and thermal deviation coefficient.

[0038] Among them, the pressure drop along the friction and the pressure drop at local bends jointly characterize the pressure drop of the water-cooled wall tubes; the average wall temperature and the highest wall temperature jointly characterize the wall temperature of the water-cooled wall tubes. Each hydrodynamic parameter is stored in association with the corresponding operating condition number and pipe number, thereby establishing a sample correspondence between the operating condition characteristic data and the hydrodynamic parameters of each water-cooled wall tube, forming a simulation sample set.

[0039] The working condition characteristic data and hydrodynamic parameters in the simulation sample set are processed by Min-Max normalization, and the input and output data are mapped to the interval [0, 1].

[0040] In one specific embodiment, 1200 sets of simulation samples are divided into a training set, a validation set, and a test set in a ratio of 7:2:1. The training set includes 840 sets of simulation samples, the validation set includes 240 sets of simulation samples, and the test set includes 120 sets of simulation samples.

[0041] The training set is used to adjust the network weights and bias parameters of the BP neural network, the validation set is used to monitor the loss changes of the model on data that are not involved in the current parameter update and to determine the training termination condition, and the test set is used to verify the prediction accuracy of the hydrodynamic characteristic prediction model after training is completed.

[0042] S102 uses operating condition characteristic data as input to the BP neural network and hydrodynamic parameters of each water-cooled wall tube as output to construct a multi-output BP neural network.

[0043] In one specific embodiment, the topology of the multi-output BP neural network includes: The input layer consists of 28 neurons, each corresponding to one of the 28 dimensions of working condition feature data. The first hidden layer consists of 120 neurons and uses the ReLU activation function; The second hidden layer consists of 72 neurons and uses the ReLU activation function; The output layer consists of 12 neurons, each corresponding to one of the 12 single-tube hydrodynamic parameters, and uses the Sigmoid activation function.

[0044] For boilers with multiple water-cooled wall tubes, the hydrodynamic characteristic prediction model can be called up one by one according to the pipe number. Alternatively, the operating condition characteristic data of different water-cooled wall tubes can be grouped into batch input data according to a predetermined data arrangement to obtain the hydrodynamic parameters of all water-cooled wall tubes in batches.

[0045] A composite loss function is constructed based on the prediction error loss, mass conservation loss, and energy conservation loss. The composite loss function is expressed as: Total Loss = α × LossMSE + β × Loss mass + γ × Loss energy

[0046] Wherein, LossMSE represents the mean square error between the BP neural network prediction and the hydrodynamic parameters output by the CFD numerical simulation model; Loss mass represents the loss caused by the inlet and outlet mass flow rates of a single water-cooled wall tube not satisfying the mass conservation relationship, used to constrain the consistency between the inflow mass and the outflow mass; Loss energy represents the loss caused by the non-satisfaction of the energy conservation relationship between the heat flow from the flue gas side, the enthalpy rise of the working fluid, and the heat dissipation from the tube wall, used to constrain the balance between the heat flow from the flue gas side, the heat absorption of the working fluid, and the heat dissipation from the tube wall.

[0047] In one specific embodiment, the weight α for prediction error loss is set to 0.7, the weight β for mass conservation loss is set to 0.18, and the weight γ for energy conservation loss is set to 0.12.

[0048] By introducing mass and energy conservation constraints during the training process of the BP neural network, the working fluid mass flow rate, working fluid enthalpy, tube wall temperature and related hydrodynamic parameters output by the network satisfy the physical relationship of flow and heat transfer inside the water-cooled wall tube, reducing the possibility of non-physical prediction results when fitting based solely on data errors.

[0049] In other embodiments, in addition to retaining the prediction error loss, mass conservation loss, and energy conservation loss, it is also possible to add the pipe wall heat flux conservation loss or pressure drop gradient conservation loss to further constrain the corresponding prediction parameters.

[0050] The multi-output backpropagation neural network was trained using the Adam optimizer, with an initial learning rate of 0.001, a learning rate decay factor of 0.96, a batch size of 32, and a maximum number of iterations (Epoch) of 1000.

[0051] During training, the composite loss function is calculated using the training set to update the network weights and bias parameters, while the validation set is used to monitor the network's generalization performance. If the validation set loss does not decrease for 30 consecutive epochs, an early stopping condition is triggered, terminating the current training and preventing overfitting from continued training.

[0052] After training, the BP neural network was validated using 120 test set samples, and the trained network weights, bias parameters, and normalized mapping parameters were saved to form a hydrodynamic characteristic prediction model and the corresponding model parameter package.

[0053] S103, in the unit online monitoring mode, acquires the operating data corresponding to the test condition through the boiler distributed control system, and forms the operating condition characteristic data of the test condition based on the acquired operating data.

[0054] The operating condition characteristic data of the test condition can include boiler load, feedwater parameters, fuel parameters, air volume parameters, coal quality parameters, furnace heat load parameters, and water-cooled wall inlet parameters. When using the above 28-dimensional input structure, 28-dimensional features with the same meaning and the same arrangement order as in the training phase are obtained from the boiler distributed control system and pre-stored boiler structure data.

[0055] The real-time operating parameters output by the distributed boiler control system are associated with the boiler structural parameters according to the pipe number and the acquisition time. Boiler structural parameters are stored after model configuration; operating parameters are continuously updated according to the acquisition cycle of the distributed boiler control system.

[0056] The normalized mapping parameters saved during the training phase are invoked, and the working condition feature data of the test condition undergoes the same normalization process as during the training phase. The normalized working condition feature data is then input into the hydrodynamic characteristic prediction model for forward propagation to obtain the prediction results under the normalized state.

[0057] The prediction results are denormalized using normalized mapping parameters, and the correspondence between the prediction results and each water-cooled wall tube is established according to the pipe number of the water-cooled wall tube, so as to obtain the hydrodynamic parameters of each water-cooled wall tube under the test condition.

[0058] In one specific embodiment, the unit includes 600 to 1200 water-cooled wall tubes. By calling the hydrodynamic characteristic prediction model one tube at a time or in batches, 12 hydrodynamic parameters corresponding to all water-cooled wall tubes are obtained. The forward propagation time for a single operation is no more than 8 ms, and the time required to complete the prediction of one test condition is less than 10 ms.

[0059] Based on the prediction results of each water-cooled wall tube, a two-dimensional distribution cloud map of the working fluid mass flow rate, tube wall temperature, thermal deviation coefficient, and cavitation fraction of the water-cooled wall tube can be generated to display the parameter distribution at different furnace locations of the water-cooled wall.

[0060] S104. A safety judgment threshold library is established in advance, which stores thermal deviation threshold, flow stability threshold, backflow critical flow rate corresponding to each water-cooled wall tube, and allowable temperature corresponding to different tube materials.

[0061] The thermal deviation coefficient, flow instability judgment index, working fluid mass flow rate, and highest pipe wall temperature of each water-cooled wall tube were compared with the corresponding safety judgment threshold.

[0062] In one specific embodiment, the risk assessment criteria include: When the thermal deviation coefficient of the water-cooled wall tube is greater than 1.2, the water-cooled wall tube is identified as a high thermal deviation tube. When the flow instability index of the water-cooled wall tube is greater than 1.0, it is determined that the water-cooled wall tube has a risk of two-phase flow pulsation. When the working fluid mass flow rate of the water-cooled wall tube is less than the corresponding backflow critical flow rate of the water-cooled wall tube, it is determined that there is a risk of working fluid stagnation or backflow in the water-cooled wall tube. When the maximum temperature of the water-cooled wall tube exceeds the allowable temperature of the corresponding tube material, it is determined that the water-cooled wall tube has a risk of local overheating.

[0063] Based on the comparison results of various hydrodynamic parameters and corresponding thresholds, water-cooled wall tubes with operational risks are identified, and the risk type, risk level and safety margin are determined according to the pre-configured risk classification rules.

[0064] The risk assessment results are output in the form of a structured report. The structured report includes at least the hazardous pipeline number, the location of the hazardous pipeline in the furnace, the risk type, the risk level, and the safety margin. Risk types include high-temperature deviation risk, two-phase flow pulsation risk, flow stagnation or backflow risk, and localized overheating risk.

[0065] In other embodiments, based on the above-mentioned risk assessment conditions, the average cavitation fraction of the two-phase flow or the peak value of the local heat load can be set as an auxiliary early warning indicator to increase the characterization dimension of the water-cooled wall tube operating status.

[0066] S105, acquire the field measured data under the test condition. The field measured data and the test condition characteristic data use the same test condition identifier and acquisition time.

[0067] In one specific embodiment, the pipe wall temperature, total loop pressure difference, and total water supply flow rate are simultaneously acquired under the same test condition and at the same acquisition time, and used as on-site measured data.

[0068] Since the hydrodynamic characteristic prediction model outputs the hydrodynamic parameters of each water-cooled wall tube, while some field measurement data are loop-level or boiler-wide data, the hydrodynamic parameters of each water-cooled wall tube are screened or summarized according to the pipe location or loop range corresponding to the field measurement point to obtain the prediction data corresponding to the field measurement data.

[0069] For example, select the predicted wall temperature of the water-cooled wall tube corresponding to the on-site pipe wall temperature measurement point; obtain the predicted data corresponding to the total pressure difference of the circuit based on the pressure drop of each water-cooled wall tube belonging to the same circuit; obtain the predicted data corresponding to the total feedwater flow rate based on the working fluid mass flow rate of the corresponding water-cooled wall tube.

[0070] The deviation between the predicted data and the field measured data is calculated, and the deviation is used as the basis for feedback correction of the CFD numerical simulation model. Based on the heat transfer or flow deviation reflected by different measured parameters, at least one of the following model parameters in the CFD numerical simulation model is corrected: furnace radiation heat transfer coefficient, water-cooled wall two-phase flow boiling heat transfer coefficient, and wall heat flux scaling coefficient.

[0071] After the model parameters are corrected, the corrected CFD numerical simulation model is used to solve the supplementary working conditions. In one specific embodiment, each feedback update generates 5 to 20 sets of supplementary simulation samples.

[0072] Supplementary simulation samples are added to the original simulation sample set to form an updated simulation sample set. This updated set is then used to perform secondary training or parameter fine-tuning of the hydrodynamic characteristic prediction model. After training, the updated network weights and bias parameters replace the corresponding parameters in the original hydrodynamic characteristic prediction model, enabling the updated model to be used for determining the hydrodynamic parameters for subsequent test conditions.

[0073] This creates the following data loop: The CFD numerical simulation model outputs a simulation sample set; the simulation sample set is used to train the hydrodynamic characteristic prediction model; the hydrodynamic characteristic prediction model outputs the hydrodynamic parameters of each water-cooled wall tube; the hydrodynamic parameters are used for risk assessment and to generate prediction data corresponding to the field measured data; the deviation between the prediction data and the field measured data is used to correct the CFD numerical simulation model; the corrected CFD numerical simulation model generates supplementary simulation samples; the supplementary simulation samples are used to update the hydrodynamic characteristic prediction model.

[0074] Example 2: After completing the construction of the CFD numerical simulation model and the training of the hydrodynamic characteristic prediction model, the hydrodynamic characteristic prediction model can also be used for offline water-cooled wall design verification during the boiler design phase.

[0075] Offline water-cooled wall design verification does not rely on real-time data collected by the boiler's distributed control system. For the water-cooled wall design schemes to be compared, design load, design coal quality, feedwater parameters, and boiler structural parameters are obtained, and corresponding operating characteristic data are generated based on different pipe diameters, pipe wall thicknesses, and loop layouts.

[0076] After normalizing the operating characteristic data of each design scheme, the data are input into the hydrodynamic characteristic prediction model to determine the hydrodynamic parameters of each water-cooled wall tube under different design schemes in batches. The different design schemes are compared based on the thermal deviation coefficient, flow instability index, working fluid mass flow rate, and maximum tube wall temperature of each water-cooled wall tube.

[0077] In practical applications, genetic algorithms or NSGA-II multi-objective optimization algorithms can be used to assist in finding the optimal solution for different water-cooled wall pipe diameters and loop arrangements. The hydrodynamic parameters output by the hydrodynamic characteristic prediction model are used as evaluation data to compare the thermal deviation and flow stability of different schemes.

[0078] By using a hydrodynamic characteristic prediction model to perform batch calculations of different design schemes, it is possible to shorten the verification cycle of water-cooled wall structure schemes without having to perform long-term three-dimensional CFD numerical iterations for each scheme.

[0079] Example 3: See Figure 2 This embodiment provides a system for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling, including a simulation sample construction module, a prediction model training module, a hydrodynamic parameter determination module, a risk assessment module, and a feedback update module.

[0080] The simulation sample construction module is used to obtain boiler structural parameters and operating parameters under different working conditions, construct a CFD numerical simulation model that couples the flue gas side of the furnace with the working fluid side inside the water-cooled wall tubes, and generate a simulation sample set using the CFD numerical simulation model.

[0081] The simulation sample construction module includes a computational domain construction unit, a batch solution unit, and a sample extraction unit.

[0082] The computational domain construction unit establishes the computational domain on the flue gas side of the furnace and the computational domain on the working fluid side of each water-cooled wall tube based on the boiler structural parameters, and establishes the heat flow data transfer relationship between the computational domain on the flue gas side of the furnace and the computational domain on the working fluid side.

[0083] The batch solution unit receives operating parameters under different working conditions, sets the operating parameters as the boundary conditions of the CFD numerical simulation model, performs coupled solution for different working conditions, and transmits the solution results to the sample extraction unit.

[0084] The sample extraction unit extracts operating condition feature data and hydrodynamic parameters of each water-cooled wall tube from the solution results, forms a simulation sample set according to the operating condition number and the pipe number, and transmits the simulation sample set to the prediction model training module.

[0085] The prediction model training module receives the simulation sample set output by the simulation sample construction module, normalizes and divides the simulation sample set, and trains the BP neural network with the operating condition feature data as input and the hydrodynamic parameters of each water-cooled wall tube as output.

[0086] The prediction model training module calculates the composite loss based on the prediction error loss, mass conservation loss, and energy conservation loss. Based on the composite loss, it updates the network weights and bias parameters of the BP neural network to obtain the hydrodynamic characteristic prediction model.

[0087] After training is completed, the prediction model training module transmits the network weights, bias parameters, and normalized mapping parameters of the hydrodynamic characteristic prediction model to the hydrodynamic parameter determination module.

[0088] The hydrodynamic parameter determination module is connected to the boiler distributed control system to receive the operating data of the test condition and call the pre-stored boiler structural parameters to form the operating condition characteristic data of the test condition.

[0089] The hydrodynamic parameter determination module normalizes the operating condition feature data based on the normalized mapping parameters transmitted by the prediction model training module, performs forward propagation of the BP neural network according to the network weights and bias parameters, and performs inverse normalization on the prediction results to obtain the hydrodynamic parameters of each water-cooled wall tube.

[0090] The hydrodynamic parameter determination module stores the hydrodynamic parameters according to the pipe number of the water-cooled wall pipe, and transmits the hydrodynamic parameters of each water-cooled wall pipe to the risk assessment module and the feedback update module respectively.

[0091] The risk assessment module receives the hydrodynamic parameters of each water-cooled wall tube output by the hydrodynamic parameter determination module and calls the pre-established safety assessment threshold library.

[0092] The risk assessment module compares the thermal deviation coefficient, flow instability assessment index, working fluid mass flow rate, and highest pipe wall temperature of each water-cooled wall tube with the corresponding thresholds to identify water-cooled wall tubes that have risks of high thermal deviation, two-phase flow pulsation, flow stagnation, backflow, or local overheating.

[0093] The risk assessment module generates risk assessment results that include the number of dangerous pipelines, furnace location, risk type, risk level, and safety margin. It can also generate a two-dimensional distribution cloud map of hydrodynamic parameters based on the prediction results of each water-cooled wall tube.

[0094] The feedback update module is connected to the boiler distributed control system, the hydrodynamic parameter determination module, the simulation sample construction module, and the prediction model training module.

[0095] The feedback update module receives the field-measured data output from the boiler distributed control system and the hydrodynamic parameters of each water-cooled wall tube output from the hydrodynamic parameter determination module. The feedback update module filters or summarizes the hydrodynamic parameters according to the measurement point location and loop relationship, obtaining predicted data corresponding to the field-measured data, and calculates the deviation between the predicted data and the field-measured data.

[0096] The feedback update module generates model parameter correction data based on the deviation and transmits the model parameter correction data to the simulation sample construction module. The simulation sample construction module modifies the furnace radiation heat transfer coefficient, the water-cooled wall two-phase flow boiling heat transfer coefficient, or the wall heat flux scaling factor based on the model parameter correction data, and generates supplementary simulation samples using the corrected CFD numerical simulation model.

[0097] The simulation sample construction module transmits supplementary simulation samples to the prediction model training module. The prediction model training module uses the supplementary simulation samples to update the hydrodynamic characteristic prediction model and transmits the updated hydrodynamic characteristic prediction model to the hydrodynamic parameter determination module for use in determining the hydrodynamic characteristics of subsequent test conditions.

[0098] Example 4: The hydrodynamic characteristic prediction model was validated using 120 sets of CFD numerical simulation conditions that were not involved in model training as a test set.

[0099] Under the test conditions of this embodiment, the average relative error of each hydrodynamic parameter under all operating conditions is no greater than 2.7%, and the prediction error under the 20% load limit condition is no greater than 4.2%.

[0100] Under the same test conditions, the calculation error of the one-dimensional analytical model is 18% to 26%; the prediction error of the BP neural network without mass conservation loss and energy conservation loss under extreme conditions is 29% to 35%; the time required for the pure CFD numerical simulation model to complete the single-condition calculation is 32h to 72h; the time required for the hydrodynamic characteristic prediction model in this embodiment to complete a single prediction is less than 10ms.

[0101] The above verification results show that by generating simulation samples covering both normal and extreme operating conditions through CFD numerical simulation models, and by introducing mass and energy conservation constraints during the training of BP neural networks, the prediction accuracy of the hydrodynamic parameters of each water-cooled wall tube can be maintained while reducing the amount of repetitive CFD calculations.

[0102] Meanwhile, by feeding back the deviation between the measured data and the corresponding predicted data to the CFD numerical simulation model, and using the corrected CFD numerical simulation model to generate supplementary simulation samples, the hydrodynamic characteristic prediction model can be continuously updated to adapt to changes in boiler operating conditions and heat exchange conditions.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling, characterized in that, Includes the following steps: Obtain boiler structural parameters and operating parameters under different working conditions, construct a CFD numerical simulation model that couples the flue gas side of the furnace with the working fluid side inside the water-cooled wall tubes, and use the CFD numerical simulation model to generate a simulation sample set containing operating condition characteristic data and hydrodynamic parameters of each water-cooled wall tube. Using operating condition characteristic data as input and hydrodynamic parameters of each water-cooled wall tube as output, a BP neural network is trained to obtain a hydrodynamic characteristic prediction model. The working condition characteristic data of the test condition are obtained and input into the hydrodynamic characteristic prediction model to obtain the hydrodynamic parameters of each water-cooled wall tube; Based on hydrodynamic parameters and preset safety judgment conditions, identify water-cooled wall tubes with operational risks and output the risk judgment results; Obtain on-site measured data of the working condition to be tested, correct the CFD numerical simulation model based on the deviation between the on-site measured data and the corresponding hydrodynamic parameters, generate supplementary simulation samples using the corrected CFD numerical simulation model, and update the hydrodynamic characteristic prediction model.

2. The method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling according to claim 1, characterized in that, The specific method for obtaining boiler structural parameters and operating parameters under different operating conditions, constructing a CFD numerical simulation model coupling the flue gas side of the furnace and the working fluid side inside the water-cooled wall tubes, and generating a simulation sample set containing operating condition characteristic data and hydrodynamic parameters of each water-cooled wall tube using the CFD numerical simulation model is as follows: Based on the boiler structural parameters, a computational domain for the flue gas side of the furnace and a computational domain for the working fluid side of each water-cooled wall tube are established. The computational domains for the flue gas side and the working fluid side are divided into multi-scale grids, and heat flow data is transferred through the interface. The flue gas flow and radiation heat transfer processes in the computational domain on the flue gas side of the furnace, as well as the working fluid flow, heat transfer, and phase change processes in the computational domain on the working fluid side, are solved in a coupled manner. Different operating conditions are set according to boiler load, feedwater parameters, fuel parameters, air volume parameters, coal quality parameters and combustion deviation parameters, and CFD numerical simulation models are used to solve the different operating conditions respectively. The operating condition characteristic data corresponding to each operating condition are extracted from the solution results, as well as the working fluid mass flow rate, pressure drop, cavitation fraction, outlet working fluid enthalpy, tube wall temperature, local heat load, critical heat flux density, flow instability judgment index, backflow critical flow rate and thermal deviation coefficient corresponding to each water-cooled wall tube, to form a simulation sample set.

3. The method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling according to claim 1, characterized in that, The specific method for training a BP neural network to obtain a hydrodynamic characteristic prediction model, using operating condition characteristic data as input and hydrodynamic parameters of each water-cooled wall tube as output, is as follows: The working condition characteristic data and hydrodynamic parameters in the simulation sample set are normalized, and the normalized simulation sample set is divided into training set, validation set and test set. A multi-output BP neural network is constructed, consisting of an input layer, at least one hidden layer, and an output layer. The neurons in the input layer correspond to the operating condition feature data, and the neurons in the output layer correspond to the hydrodynamic parameters of each water-cooled wall tube. A composite loss function is constructed based on prediction error loss, mass conservation loss, and energy conservation loss. The multi-output BP neural network is trained using the training set and the composite loss function. The training termination condition is determined using the validation set, and the training results are verified using the test set. Save the network weights, bias parameters, and normalized mapping parameters of the trained multi-output BP neural network to obtain the hydrodynamic characteristic prediction model.

4. The method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling according to claim 1, characterized in that, The specific method for obtaining the operating condition characteristic data of the test condition and inputting it into the hydrodynamic characteristic prediction model to obtain the hydrodynamic parameters of each water-cooled wall tube is as follows: Acquire at least one of the following parameters for the operating condition under test: boiler load, feedwater parameters, fuel parameters, air volume parameters, coal quality parameters, furnace heat load parameters, and water-cooled wall inlet parameters, to form operating condition characteristic data for the operating condition under test. The working condition characteristic data of the test condition are normalized according to the normalized mapping parameters, and the normalized working condition characteristic data are input into the hydrodynamic characteristic prediction model for forward propagation. The prediction results output by the hydrodynamic characteristic prediction model are inversely normalized. The correspondence between the prediction results and each water-cooled wall tube is established according to the pipe number of the water-cooled wall tube, and the hydrodynamic parameters of each water-cooled wall tube are obtained.

5. The method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling according to claim 1, characterized in that, The specific method for identifying water-cooled wall tubes with operational risks and outputting risk assessment results based on hydrodynamic parameters and preset safety assessment conditions is as follows: The thermal deviation coefficient, flow instability judgment index, working fluid mass flow rate and maximum pipe wall temperature of each water-cooled wall tube are compared with the corresponding safety judgment threshold. When the thermal deviation coefficient of the water-cooled wall tube exceeds the thermal deviation threshold, it is determined that the water-cooled wall tube has a high risk of thermal deviation. When the flow instability determination index of the water-cooled wall tube exceeds the flow stability threshold, it is determined that the water-cooled wall tube has a risk of two-phase flow pulsation. When the working fluid mass flow rate of the water-cooled wall tube is lower than the corresponding critical backflow flow rate, it is determined that there is a risk of flow stagnation or backflow in the water-cooled wall tube. When the maximum wall temperature of the water-cooled wall tube exceeds the allowable temperature of the corresponding tube material, it is determined that there is a risk of local overheating in the water-cooled wall tube. Based on the comparison results, determine the risk type, risk level, and safety margin of each water-cooled wall tube, and output the risk assessment results including the dangerous pipe number, furnace location, risk type, and safety margin.

6. The method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling according to claim 1, characterized in that, The specific method for obtaining on-site measured data of the working condition to be tested, correcting the CFD numerical simulation model based on the deviation between the on-site measured data and the corresponding hydrodynamic parameters, and using the corrected CFD numerical simulation model to generate supplementary simulation samples to update the hydrodynamic characteristic prediction model is as follows: The pipe wall temperature, total loop pressure difference, and total water supply flow rate under the same test condition and at the same acquisition time are used as field measured data. The hydrodynamic parameters of each water-cooled wall tube are screened or summarized to obtain predicted data corresponding to the field measured data, and the deviation between the predicted data and the field measured data is calculated. Correct at least one of the following model parameters in the CFD numerical simulation model based on the deviation: furnace radiation heat transfer coefficient, water-cooled wall two-phase flow boiling heat transfer coefficient, and wall heat flux scaling coefficient. The modified CFD numerical simulation model is used to solve the supplementary working conditions, generating supplementary simulation samples. These supplementary simulation samples are then added to the simulation sample set, and the updated simulation sample set is used to update the hydrodynamic characteristic prediction model.

7. A system for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling, characterized in that, include: The simulation sample construction module is used to obtain boiler structural parameters and operating parameters under different working conditions, construct a CFD numerical simulation model that couples the flue gas side of the furnace with the working fluid side inside the water-cooled wall tubes, and use the CFD numerical simulation model to generate a simulation sample set containing operating condition characteristic data and hydrodynamic parameters of each water-cooled wall tube. The prediction model training module is used to train the BP neural network with the operating condition feature data as input and the hydrodynamic parameters of each water-cooled wall tube as output, so as to obtain the hydrodynamic characteristic prediction model. The hydrodynamic parameter determination module is used to acquire the working condition characteristic data of the working condition to be tested and input it into the hydrodynamic characteristic prediction model to obtain the hydrodynamic parameters of each water-cooled wall tube. The risk assessment module is used to identify water-cooled wall tubes with operational risks based on hydrodynamic parameters and preset safety assessment conditions, and output the risk assessment results. The feedback update module is used to acquire the field measured data of the working condition to be tested, correct the CFD numerical simulation model based on the deviation between the field measured data and the corresponding hydrodynamic parameters, generate supplementary simulation samples using the corrected CFD numerical simulation model, and update the hydrodynamic characteristic prediction model.

8. The boiler water-cooled wall hydrodynamic characteristic determination system based on CFD-BP coupling according to claim 7, characterized in that, The simulation sample construction module includes: The computational domain construction unit is used to establish the computational domain on the flue gas side of the furnace and the computational domain on the working fluid side of each water-cooled wall tube based on the boiler structural parameters, and to establish the heat flow data transfer relationship between the computational domain on the flue gas side of the furnace and the computational domain on the working fluid side. The batch solution unit is used to receive operating parameters under different working conditions, set the operating parameters as boundary conditions of the CFD numerical simulation model, and perform coupled solution for different working conditions. The sample extraction unit is used to extract operating condition feature data and hydrodynamic parameters of each water-cooled wall tube from the solution results output by the batch solution unit, form a simulation sample set, and transmit the simulation sample set to the prediction model training module.

9. The method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling according to claim 7, characterized in that: The prediction model training module is also used to transmit the network weights, bias parameters, and normalized mapping parameters of the hydrodynamic characteristic prediction model to the hydrodynamic parameter determination module. The hydrodynamic parameter determination module is connected to the boiler distributed control system to receive the operating condition characteristic data of the operating condition to be measured, process the operating condition characteristic data according to the normalized mapping parameters, and generate the hydrodynamic parameters of each water-cooled wall tube using network weights and bias parameters. The hydrodynamic parameter determination module is also used to transmit the hydrodynamic parameters of each water-cooled wall tube to the risk assessment module, so that the risk assessment module can generate a risk assessment result that includes the dangerous pipeline number, furnace location, risk type and safety margin.

10. The method for determining the hydrodynamic characteristics of boiler water-cooled walls based on CFD-BP coupling according to claim 7, characterized in that: The feedback update module is communicatively connected to the boiler distributed control system, the hydrodynamic parameter determination module, the simulation sample construction module, and the prediction model training module, respectively. The feedback update module is used to receive the field measured data output by the boiler distributed control system and the hydrodynamic parameters output by the hydrodynamic parameter determination module, determine the predicted data corresponding to the field measured data, and calculate the deviation between the predicted data and the field measured data. The feedback update module is also used to transmit model parameter correction data to the simulation sample construction module according to the deviation, so that the simulation sample construction module can generate supplementary simulation samples using the corrected CFD numerical simulation model, and transmit the supplementary simulation samples to the prediction model training module. The prediction model training module is used to update the hydrodynamic characteristic prediction model using supplementary simulation samples, and then transmits the updated hydrodynamic characteristic prediction model to the hydrodynamic parameter determination module.

Images