Performance prediction method for fluidic screen flue gas capture system based on genetic algorithm and BP neural network

Abstract

The jet flow shielding flue gas capture system performance prediction method based on genetic algorithm fusion BP neural network belongs to the field of steel refining process ventilation and dust removal. It solves the problems of existing steel refining process flue gas capture system, such as multiple parameter interaction, obvious dynamic change and low prediction efficiency of traditional method. It includes: obtaining historical data of flue gas capture system under different working conditions and preprocessing; taking the initial weight and threshold value of BP neural network as the optimization object of genetic algorithm coding, taking the mean square error of BP neural network prediction value and actual capture efficiency value as the fitness function, introducing the fluid mechanics physical constraint penalty term, and obtaining the optimal weight and threshold value combination through genetic algorithm iteration; training the BP neural network model optimized by genetic algorithm; the trained model is used as the flue gas capture system performance prediction model, the real-time pollution source and the air supply and exhaust parameters are input, and the capture efficiency prediction value is output. It is used for steel refining process.

Description

Technical Field

[0001] This invention belongs to the field of ventilation and dust removal technology in steel refining processes, and in particular relates to a performance prediction method for jet shielded flue gas capture systems based on genetic algorithms and BP neural networks. Background Technology

[0002] In steel refining processes, such as LF furnaces and RH vacuum treatment, molten steel is exposed to air and undergoes vigorous chemical reactions, generating large amounts of high-temperature flue gas carrying fine particulate matter. These flue gases are extremely hot, typically reaching hundreds or even thousands of degrees Celsius, and due to strong thermal buoyancy, their diffusion behavior is significantly disordered and dynamic. If not properly controlled, the flue gas will permeate large areas of the workshop, severely deteriorating the working environment, reducing visibility, and affecting production safety, and posing long-term health hazards to the respiratory system of on-site operators. Therefore, an efficient and reliable flue gas capture system is an indispensable key element in clean steel metallurgical production.

[0003] Currently, local ventilation capture systems based on the jet shielding principle have been applied in this field. This system forms an air curtain by arranging slotted air inlets around the pollution source, working in conjunction with the exhaust hood above to confine and guide the disorderly spreading high-temperature flue gas into the exhaust hood. Compared to a traditional single exhaust hood, this system theoretically controls flue gas dispersion more effectively. However, its actual capture performance is a typical multi-physics field strongly coupled, unsteady dynamic process, influenced by the complex interaction of multiple parameters such as the pollution source emission rate, flue gas temperature, jet airflow, exhaust volume, and the overall airflow organization of the workshop. These parameters exhibit highly nonlinear relationships. This complexity makes system performance prediction and parameter optimization exceptionally difficult.

[0004] To evaluate and predict the performance of such systems, existing technologies mainly rely on two approaches. The first is experimental testing, which involves building a scaled-down physical model of the actual process in the laboratory and measuring the capture efficiency by changing operating parameters. While the results are relatively reliable, this method is costly, time-consuming, and limited by model size and testing conditions, making it difficult to cover the ever-changing full-condition scenarios in actual production. The limited sample data available restricts its application in rapid design and online adjustment in practical engineering. The second approach is computational fluid dynamics (CFD) numerical simulation. This method, by solving the Navier-Stokes equations, can reveal the distribution patterns of flow, temperature, and concentration fields in detail, and has become an important tool for system mechanism research and design. However, facing the need for optimization of massive operating conditions with multiple combinations of variables such as pollution source intensity, temperature, supply air volume, and exhaust air volume, each CFD simulation requires a large amount of computational resources and time, often measured in hours or even days. This is completely unable to meet the real-time requirements of industrial sites for rapid prediction and dynamic parameter adjustment at the second or minute level when operating conditions change abruptly.

[0005] Due to the aforementioned highly dynamic and strongly coupled complexity, those skilled in the art generally recognize that the steel refining flue gas capture process cannot achieve high-precision real-time prediction using a single, historically data-driven mathematical model such as a traditional neural network. Therefore, current engineering practice and research are still mainly limited to offline analysis using the aforementioned time-consuming CFD simulations, or to rough estimation and adjustment relying on empirical formulas with narrow applicability derived from limited experimental data. This approach has fundamental flaws: low prediction efficiency, inability to respond to real-time changing operating conditions; and a sharp drop in prediction accuracy and insufficient reliability under complex operating conditions exceeding the range of empirical formulas. Ultimately, this leads to the design and operating parameter matching of flue gas capture systems often relying on trial and error, making it difficult to achieve the synergistic optimization goal of system energy saving (avoiding over-ventilation) while ensuring efficient capture. Summary of the Invention

[0006] In view of this, the present invention aims to propose a performance prediction method for jet shielded flue gas capture system based on genetic algorithm and BP neural network, so as to solve the problems of multi-parameter interaction, obvious dynamic changes and low prediction efficiency of traditional methods in the existing flue gas capture system of steel refining process.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network, the method comprising: Step S1: Obtain historical data of the flue gas capture system operating under different conditions. The historical data includes feature variables and corresponding true values ​​of capture efficiency. The feature variables include pollution source parameters and supply and exhaust air parameters. The historical data is preprocessed and divided into training set and test set. Step S2: Construct a BP neural network prediction model, including: determining the number of nodes in the input layer, hidden layer, and output layer of the BP neural network, and setting the network activation function and training parameters; Step S3: Construct a BP neural network model optimized by a genetic algorithm, including: encoding the initial weights and thresholds of the BP neural network as the optimization objects of the genetic algorithm, using the mean square error between the predicted value and the actual trapping efficiency value of the BP neural network as the fitness function, introducing a fluid dynamics physical constraint penalty term, and obtaining the optimal combination of weights and thresholds through iterative operations of the genetic algorithm. Step S4: Use the training set to train the BP neural network model optimized by the genetic algorithm, use the test set to verify the trained model, and use the root mean square error as the loss function to evaluate the model performance. Step S5: Use the trained model as the final flue gas capture system performance prediction model, input the real-time pollution source parameters and supply and exhaust air parameters, and output the corresponding capture efficiency prediction value.

[0008] Furthermore, a preferred method is proposed, in which the characteristic variables in step S1 specifically include: pollution source diameter, pollutant emission rate, pollutant emission temperature, jet air volume, and exhaust air volume.

[0009] Furthermore, a preferred approach is proposed, wherein step S1 further includes: constructing dimensionless derived features based on the feature variables as model input; the dimensionless derived features include Archimedes numbers used to characterize the intensity of pollution sources. And the ejector ratio, used to measure the impact of jet flow rate and exhaust flow rate on the capture efficiency of local exhaust hoods. ε .

[0010] Furthermore, a preferred method is proposed, wherein the Archimedes number for:

[0011] in, g It is the acceleration due to gravity; d The characteristic length of the pollution source; U cs For the speed of pollution sources; T cs The absolute temperature of the pollution source; T r For reference absolute temperature.

[0012] Furthermore, a preferred method is proposed, wherein the number of nodes l in the hidden layer is determined according to the following empirical formula:

[0013] Where p is the number of neurons in the input layer, m is the number of neurons in the output layer, and a is an integer between [1, 10].

[0014] Furthermore, a preferred embodiment is proposed, wherein in step S3, the fluid dynamics physical constraint penalty term includes: When the exhaust volume parameter does not reach the preset threshold, individuals that cause the predicted capture efficiency value to be higher than the baseline value are subject to fitness penalty. When the Archimedes number exceeds a preset threshold, individuals whose predicted capture efficiency does not decrease with increasing Archimedes number are subject to fitness penalties.

[0015] Furthermore, a preferred method is proposed, in which the parameters of the genetic algorithm in step S3 are set as follows: population size of 20, crossover probability of 0.6, mutation probability of 0.1, and maximum number of generations of evolution of 50.

[0016] Furthermore, a preferred method is proposed, wherein the true value of the capture efficiency in the historical data covers the invalid interval, the inefficient interval, the medium-efficiency interval, and the high-efficiency interval; wherein the invalid interval is η < 70%, the inefficient interval is 70% ≤ η < 80%, the medium-efficiency interval is 80% ≤ η < 90%, and the high-efficiency interval is η > 90%; where η is the capture efficiency.

[0017] Based on the same inventive concept, the present invention also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the performance prediction method for a jet shielded flue gas capture system based on a genetic algorithm fused with a BP neural network as described in any of the preceding claims.

[0018] Based on the same inventive concept, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network as described above.

[0019] Compared with the prior art, the beneficial effects of the present invention are: Existing technologies such as CFD simulation are essentially based on numerical solutions to physical laws. Their prediction accuracy depends on precise boundary conditions, turbulence models, and huge computational grids and time steps, resulting in high computational costs. Traditional empirical formulas, based on static mathematical relationships fitted from limited experimental data, cannot characterize the complex nonlinear interactions and dynamic processes between multiple parameters. The core of this invention is to construct a surrogate model that can accurately map the complex nonlinear relationship between system input parameters and capture efficiency output. Specifically, it utilizes the powerful nonlinear fitting capability of a BP neural network as a basic function approximator for this mapping relationship. Through its multi-layered hidden layer structure, the neural network can automatically learn and internalize the implicit, high-dimensional nonlinear laws between pollution source parameters, supply and exhaust air parameters, and the final capture efficiency—laws that are difficult to express using traditional explicit mathematical formulas. Furthermore, a genetic algorithm is introduced to globally optimize the weights and thresholds of the neural network, which is key to solving the technical challenges. Traditional BP neural network training uses gradient descent, whose performance is highly dependent on the random setting of initial weights and thresholds, making it prone to getting trapped in local optima, leading to unstable model predictions and limited accuracy. This invention combines the global search capability of a genetic algorithm with the local fine-grained search capability of a backpropagation (BP) neural network. The genetic algorithm encodes weights and thresholds as chromosomes, uses network prediction error as fitness, and performs a global parallel search in the solution space by simulating selection, crossover, and mutation mechanisms in biological evolution, thereby finding a better set of initial parameters for the BP network. This effectively points the neural network training to a starting point closer to the global optimum, ensuring that the final trained model has higher prediction accuracy and better convergence stability.

[0020] This invention addresses the challenge of insufficient prediction reliability of a single BP neural network in complex industrial scenarios. By employing a pre-optimization global algorithm using a genetic algorithm, this invention effectively avoids this pitfall, improving the model's robustness and the reliability of prediction results. Furthermore, this invention introduces fluid dynamics physical constraints as a penalty term into the fitness function of the genetic algorithm. For example, it penalizes predictions that contradict physical principles, such as abnormally high capture efficiency when exhaust volume is insufficient or the jet effect not weakening when buoyancy is excessive, thus guiding the optimization process. Simultaneously, while CFD methods offer acceptable accuracy but are slow, and empirical formulas are fast but inaccurate, this invention uses time-consuming, high-fidelity CFD simulations or experimental data as offline training samples to train and optimize the GA-BP model. Once the model training is complete, the computation time for a single prediction during the application phase is extremely short, achieving offline training and online prediction within seconds, thereby meeting the needs of real-time performance evaluation and rapid parameter matching in industrial settings. The BP neural network model optimized by genetic algorithm and incorporating physical constraints can achieve a determination coefficient of over 0.985 between its predicted and actual values, possessing prediction accuracy close to or equivalent to that of high-fidelity CFD simulation, far superior to traditional empirical formulas.

[0021] After the model training of this invention is completed, the performance prediction of any new operating condition can be completed in seconds or even milliseconds. Compared with CFD simulation that requires several hours or even days, the prediction efficiency is improved by several orders of magnitude, making it possible for the first time to perform real-time performance evaluation and online parameter tuning of complex flue gas capture systems.

[0022] The method proposed in this invention can train a predictive model covering the entire range of operating conditions from ineffective, inefficient, medium-efficient to high-efficiency using existing data, and based on this, quickly generate a performance prediction scheme library covering a large number of parameter combinations. This provides real-time, quantitative data support for equipment selection and parameter matching in the early stages of system design, as well as dynamic optimization and adjustment during operation, overcoming the blindness and limitations of traditional trial-and-error methods or finite-condition simulations. Attached Figure Description

[0023] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of the flue gas capture system performance prediction method described in this invention; Figure 2 This is a flowchart of the GA-BP algorithm optimization described in this invention; Figure 3 This is a comparison chart of the predicted and actual values ​​of the GA-BP neural network described in this invention. Figure 4 This is a graph showing the deviation between the predicted and actual values ​​of the GA-BP neural network described in this invention. Figure 5 This is a schematic diagram illustrating the prediction results under different combinations of pollution source intensity and supply / exhaust air ratio as described in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other, and the described embodiments are only some embodiments of the present invention, not all embodiments.

[0025] Implementation Method 1: This implementation method addresses the problems of multi-parameter interaction, significant dynamic changes, and low prediction efficiency of traditional methods in existing steel refining process flue gas capture systems. It proposes a performance prediction method for jet-shielded flue gas capture systems based on a genetic algorithm fused with a BP neural network. The method includes: Step S1: Obtain historical data of the flue gas capture system operating under different conditions. The historical data includes feature variables and corresponding true values ​​of capture efficiency. The feature variables include pollution source parameters and supply and exhaust air parameters. The historical data is preprocessed and divided into training set and test set. Step S2: Construct a BP neural network prediction model, including: determining the number of nodes in the input layer, hidden layer, and output layer of the BP neural network, and setting the network activation function and training parameters; Step S3: Construct a BP neural network model optimized by a genetic algorithm, including: encoding the initial weights and thresholds of the BP neural network as the optimization objects of the genetic algorithm, using the mean square error between the predicted value and the actual trapping efficiency value of the BP neural network as the fitness function, introducing a fluid dynamics physical constraint penalty term, and obtaining the optimal combination of weights and thresholds through iterative operations of the genetic algorithm. Step S4: Use the training set to train the BP neural network model optimized by the genetic algorithm, use the test set to verify the trained model, and use the root mean square error as the loss function to evaluate the model performance. Step S5: Use the trained model as the final flue gas capture system performance prediction model, input the real-time pollution source parameters and supply and exhaust air parameters, and output the corresponding capture efficiency prediction value.

[0026] In step S1 of this embodiment, the characteristic variables specifically include: pollution source diameter, pollutant emission rate, pollutant emission temperature, jet air volume, and exhaust air volume.

[0027] Furthermore, step S1 also includes: constructing dimensionless derived features based on the feature variables as model input; the dimensionless derived features include Archimedes numbers used to characterize the intensity of pollution sources. And the ejector ratio, used to measure the impact of jet flow rate and exhaust flow rate on the capture efficiency of local exhaust hoods. ε .

[0028] Furthermore, the Archimedes number for:

[0029] in, g It is the acceleration due to gravity; d The characteristic length of the pollution source; U cs For the speed of pollution sources; T cs The absolute temperature of the pollution source; T r For reference absolute temperature.

[0030] The number of nodes l in the hidden layer described in this embodiment is determined according to the following empirical formula:

[0031] Where p is the number of neurons in the input layer, m is the number of neurons in the output layer, and a is an integer between [1, 10].

[0032] In step S3 of this embodiment, the fluid dynamics physical constraint penalty term includes: When the exhaust volume parameter does not reach the preset threshold, individuals that cause the predicted capture efficiency value to be higher than the baseline value are subject to fitness penalty. When the Archimedes number exceeds a preset threshold, individuals whose predicted capture efficiency does not decrease with increasing Archimedes number are subject to fitness penalties.

[0033] Furthermore, in step S3, the parameters of the genetic algorithm are set as follows: population size of 20, crossover probability of 0.6, mutation probability of 0.1, and maximum number of generations of evolution of 50.

[0034] The true capture efficiency values ​​in the historical data described in this embodiment cover the invalid interval, the inefficient interval, the medium-efficiency interval, and the high-efficiency interval; the invalid interval is η < 70%, the inefficient interval is 70% ≤ η < 80%, the medium-efficiency interval is 80% ≤ η < 90%, and the high-efficiency interval is η > 90%; where η is the capture efficiency.

[0035] Implementation Method 2, see below Figures 1 to 5 This embodiment describes a specific implementation of the performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network, as described in Embodiment 1. The embodiment includes: Step 1: Data Acquisition and Preprocessing Historical data of the flue gas capture system under different operating conditions were obtained through scaled-down experiments and numerical simulations. This data includes the following characteristic variables: pollution source diameter, pollutant emission rate, pollutant emission temperature, jet air volume, exhaust volume, and the corresponding true value of capture efficiency. Among them, the capture efficiency is the combined capture efficiency considering the capture share of local exhaust hoods and roof exhaust hoods.

[0036] Meanwhile, the datasets used in this implementation method all contain actual samples across the entire operating condition range, covering: invalid intervals (η < 70%), inefficient intervals (70% ≤ η < 80%), medium-efficient intervals (80% ≤ η < 90%), and efficient intervals (η > 90%). This enhances the model's predictive robustness across the entire operating condition range, especially under hazardous conditions.

[0037] Furthermore, dimensionless derived features based on fluid dynamics theory are used as model inputs, significantly improving physical interpretability and prediction accuracy. These features include: the Archimedes number Ar (characterizing the relative strength of buoyancy and inertial forces) characterizing the pollution source intensity; and the ejection ratio. ε To measure the impact of jet air volume and exhaust air volume on the collection efficiency of local exhaust hoods.

[0038]

[0039] in, g Let be the acceleration due to gravity, taken as 9.81. m / s 2 ; d The characteristic length of the pollution source is in meters (m). U cs The velocity of the pollution source is m / s; T cs Let K be the absolute temperature of the pollution source. T r For reference absolute temperature, K.

[0040]

[0041] In the formula, For the ejection ratio, Q j For jet air volume, m 3 / h; Q e The exhaust volume of the exhaust hood is in meters (m). 3 / h.

[0042] Finally, the collected raw data was checked and missing and outlier values ​​were processed; data units and formats were standardized. The preprocessed complete dataset was randomly divided into a 90% training set and a 10% test set. The min-max normalization method was used to normalize the feature variables and capture efficiency values ​​in the training and test sets, linearly mapping all data to the [0,1] interval to eliminate the influence of unit dimensions and accelerate model training convergence. The transformation function is:

[0043] In the formula: X These are sample values ​​from the original data. X min The minimum value in the original data. X max The maximum value in the original data. X norm These are the normalized data values; Step 2: Construct a BP neural network prediction model A backpropagation (BP) neural network model is constructed, employing a BP neural network structure comprising an input layer, hidden layers, and an output layer. The number of nodes in the input layer is determined based on the number of factors affecting the capture efficiency. The network has p input layer nodes and m output layer nodes, corresponding to one result, i.e., the predicted capture efficiency value.

[0044] The number of nodes in the hidden layer can be determined using the following empirical formula:

[0045] In the formula, l This represents the number of neurons in the hidden layer. p The number of neurons in the input layer. m The number of neurons in the output layer. a It is an integer between [1, 10].

[0046] Furthermore, the activation functions of the network are set: the ReLU function is used for the hidden layers, and the activation function for the output layer is determined based on the range of the trapping efficiency. Since the trapping efficiency is in the [0,1] interval, the Sigmoid function is chosen. Network training parameters are set: the learning rate is typically selected between 0.01 and 0.8. By comparing the training errors under different learning rates, for example, 0.1 is chosen as the optimal value. The training target error determines the termination condition for network training, which is set to 1×10⁻⁶. -6 The number of training sessions was set to 1000.

[0047] Step 3: Construct a BP neural network model optimized by genetic algorithm (GA) First, calculate the initial weights and thresholds based on the number of nodes in the input layer, hidden layer, and output layer determined in step 2. The initial weights are calculated according to the formula... p × l + l × m The threshold is calculated as follows: the total number of hidden layer nodes plus the number of output layer nodes. l + m indivual.

[0048] In this embodiment, the BP neural network model optimized by the genetic algorithm (GA) is designed based on a jet-shielded flue gas capture system. The network's input layer has 5 nodes, corresponding to 5 feature variables: pollution source diameter, pollutant emission rate, pollutant emission temperature, jet flow rate, and exhaust flow rate. The output layer has 1 node, representing the predicted capture efficiency. The hidden layer structure was determined experimentally: one hidden layer was set, and the number of neurons in the hidden layer was adjusted within the range of 4 to 12 and trained accordingly. Based on the network's performance on the validation set, the final number of neurons in the hidden layer was determined to be 8.

[0049] All initial weights and thresholds of the BP neural network are used as the optimization objects of the genetic algorithm. All weights and thresholds are concatenated in sequence to form a single chromosome. The value of each weight and threshold is encoded in binary encoding to facilitate computer recognition. The encoding length is determined according to the number of weights and thresholds. The encoding length is equal to the sum of the total number of weights and the total number of thresholds.

[0050] Based on the mean square error (MSE) between the predicted and actual values ​​of the BP neural network, a fluid dynamics physical constraint penalty term is introduced. Simultaneously, fluid dynamics physical constraints are incorporated to constrain unreasonable prediction results that do not conform to the flow laws of high-temperature flue gas, including: (1) When the exhaust volume is insufficient, the collection efficiency will not be abnormally high; (2) When the Archimedes number is too large and the buoyancy is too strong, the jet induction effect will decrease significantly, so as to ensure that the model output conforms to the physical mechanism and engineering reality.

[0051] Next, the mean squared error (MSE) between the predicted values ​​and the actual capture efficiency values ​​of the BP neural network is used as the fitness function, and its expression is:

[0052] in, For the sample size, This represents the actual capture efficiency value. The smaller the error, the higher the fitness.

[0053] Then, the genetic operations are set up. The selection operation uses roulette wheel selection, choosing superior individuals to enter the next generation population based on their fitness ratio. Fitness and individual p... i The probability of being selected is calculated by the following formula:

[0054]

[0055] In the formula, f i Let be the fitness value of individual i. Individuals with higher fitness have a greater probability of being selected to be passed on to the next generation, thus achieving survival of the fittest; n is the number of individuals in the population.

[0056] Furthermore, individuals are encoded using real numbers, and the real number crossover method is selected. k The chromosome and the l The crossover operation of chromosomes at point j is as follows:

[0057] In the formula, b It is a random number in [0,1].

[0058] Furthermore, for the mutation operation, the first... i The first individual j One gene a ij The procedure for performing mutations is as follows:

[0059] In the formula, a max For genes a ij The upper bound; a min For genes a ij The lower bound; r It is a random number within the interval [0, 1]. f (g) Calculate according to the following formula:

[0060] r 2 is a random number; g This represents the current iteration number. G max It represents the maximum number of evolutions.

[0061] The termination condition for the genetic algorithm is set to reach the maximum number of iterations or the fitness function value reaches a preset threshold. When the termination condition is met, the genetic algorithm stops iterating and outputs the optimal combination of weights and thresholds.

[0062] In summary, the parameter settings for the genetic algorithm of the performance prediction model of the local space induced jet ventilation system based on the GA-BP neural network are shown in the table below.

[0063] Table 1 Genetic Algorithm Parameter Settings

[0064] Depend on Figure 2 As can be seen, this process demonstrates the complete steps of optimizing a BP neural network using a genetic algorithm, then assigning the optimization results to the improved BP network, further calculating errors and updating parameters, and finally outputting the prediction results after meeting the error requirements.

[0065] Step 4: Training and evaluating the neural network model optimized by the genetic algorithm The BP neural network model optimized by the genetic algorithm is trained using the training set data prepared in step 1. During training, the weights and thresholds are adjusted based on the error between the predicted and actual values ​​according to the backpropagation algorithm of the BP neural network, continuously optimizing the model. Changes in the training error are recorded during training. When the training error stops decreasing or decreases by less than a set threshold for several consecutive times, the model is considered to have converged, and training is stopped. The trained model is validated using a test set, with the root mean square error (RSME) used as the loss function.

[0066]

[0067] in, For the sample size, This represents the actual capture efficiency value. For predicted values The model's generalization ability, i.e., its predictive ability on unseen data, is evaluated using a loss function. If the model's performance metrics on the test set are unsatisfactory, such as excessively large root mean square error or excessively small coefficient of determination, further adjustments to the model structure, parameters, or optimization algorithm are needed, followed by retraining and validation until the model's performance meets the requirements. Based on the prediction results, four operating condition intervals—ineffective, inefficient, moderately efficient, and highly efficient—are automatically categorized, achieving integrated prediction and operating condition diagnosis.

[0068] Step 5: Application of Model Prediction The BP neural network model trained and optimized by the genetic algorithm (GA) in step 4 serves as the final performance prediction model for the flue gas capture device, forming a solution library. In practical applications, by inputting the parameters of the above capture system—pollution source diameter, pollutant emission rate, pollutant emission temperature, jet flow rate, and exhaust flow rate—into this model, the corresponding predicted capture efficiency value can be output, thereby achieving rapid matching, accurate prediction, and evaluation of the performance of flue gas capture devices based on air jet shielding.

[0069] Depend on Figure 3 As shown, the accuracy of the BP neural network trapping efficiency prediction model optimized by the genetic algorithm is significantly improved, and the curves of the predicted values ​​and actual values ​​are in good agreement, demonstrating good consistency. This indicates that the model's predicted values ​​highly match the true values, effectively improving the accuracy and reliability of the predictions. Figure 4 As shown, the test points corresponding to the actual and predicted values ​​of the model are mostly concentrated near the 45° regression line, exhibiting a strong linear relationship. The overall goodness of fit R² of the model reaches 0.986 (0.985 in the training phase and 0.992 in the testing phase), which is close to the ideal value of 1, indicating a high linear correlation between the predicted and actual values.

[0070] Depend on Figure 5As shown, based on the capture efficiency, it can be divided into three intervals: First, the ineffective interval, in which the jet not only fails to optimize the effect but also reduces the control effect of the exhaust hood on pollutants; second, the inefficient interval, in which the combination of supply and exhaust air parameters fails to achieve efficient capture of pollutants, possibly because the parameters are not adapted to the emission characteristics of the pollution source, resulting in easy diffusion of pollutants and insufficient capture; third, the medium-efficiency interval, in which the supply and exhaust air parameters are relatively optimized and can effectively control the diffusion and capture of pollutants to a certain extent; and fourth, the high-efficiency interval, in which the supply and exhaust air parameters are well matched with the characteristics of the pollution source, enabling rapid capture and discharge of pollutants, minimizing environmental residues, which is especially important in places with high air quality requirements.

[0071] Implementation Method 3: This implementation method proposes a computer device, including a memory and a processor. The memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes the performance prediction method for a jet shielded flue gas capture system based on a genetic algorithm fused with a BP neural network, as described in any one of Implementation Methods 1 to 2.

[0072] Implementation Method 4: This implementation method proposes a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it executes the steps of the performance prediction method for a jet shielded flue gas capture system based on a genetic algorithm fused with a BP neural network, as described in any one of Implementation Methods 1 to 2.

[0073] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0074] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit its protection scope. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading this disclosure, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the pending claims.

Claims

1. A performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network, characterized in that, The method includes: Step S1: Obtain historical data of the flue gas capture system operating under different conditions. The historical data includes feature variables and corresponding true values ​​of capture efficiency. The feature variables include pollution source parameters and supply and exhaust air parameters. The historical data is preprocessed and divided into training set and test set. Step S2: Construct a BP neural network prediction model, including: determining the number of nodes in the input layer, hidden layer, and output layer of the BP neural network, and setting the network activation function and training parameters; Step S3: Construct a BP neural network model optimized by a genetic algorithm, including: encoding the initial weights and thresholds of the BP neural network as the optimization objects of the genetic algorithm, using the mean square error between the predicted value and the actual trapping efficiency value of the BP neural network as the fitness function, introducing a fluid dynamics physical constraint penalty term, and obtaining the optimal combination of weights and thresholds through iterative operations of the genetic algorithm. Step S4: Use the training set to train the BP neural network model optimized by the genetic algorithm, use the test set to verify the trained model, and use the root mean square error as the loss function to evaluate the model performance. Step S5: Use the trained model as the final flue gas capture system performance prediction model, input the real-time pollution source parameters and supply and exhaust air parameters, and output the corresponding capture efficiency prediction value.

2. The performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network according to claim 1, characterized in that, In step S1, the characteristic variables specifically include: pollution source diameter, pollutant emission rate, pollutant emission temperature, jet air volume, and exhaust air volume.

3. The performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network according to claim 2, characterized in that, Step S1 further includes: constructing dimensionless derived features based on the feature variables as model input; the dimensionless derived features include Archimedes numbers used to characterize the intensity of pollution sources. And the ejector ratio, used to measure the impact of jet flow rate and exhaust flow rate on the capture efficiency of local exhaust hoods. ε .

4. The performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network according to claim 3, characterized in that, The Archimedes number for: in, g It is the acceleration due to gravity; d The characteristic length of the pollution source; U cs For the speed of pollution sources; T cs The absolute temperature of the pollution source; T r For reference absolute temperature.

5. The performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network according to claim 1, characterized in that, The number of nodes l in the hidden layer is determined according to the following empirical formula: Where p is the number of neurons in the input layer, m is the number of neurons in the output layer, and a is an integer between [1, 10].

6. The performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network according to claim 1, characterized in that, In step S3, the fluid dynamics physical constraint penalty term includes: When the exhaust volume parameter does not reach the preset threshold, individuals that cause the predicted capture efficiency value to be higher than the baseline value are subject to fitness penalty. When the Archimedes number exceeds a preset threshold, individuals whose predicted capture efficiency does not decrease with increasing Archimedes number are subject to fitness penalties.

7. The performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network according to claim 1, characterized in that, In step S3, the parameters of the genetic algorithm are set as follows: population size is 20, crossover probability is 0.6, mutation probability is 0.1, and maximum number of generations is 50.

8. The performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network according to claim 1, characterized in that, The true capture efficiency values ​​in the historical data cover the invalid interval, inefficient interval, medium-efficient interval, and high-efficiency interval; the invalid interval is η < 70%, the inefficient interval is 70% ≤ η < 80%, the medium-efficient interval is 80% ≤ η < 90%, and the high-efficiency interval is η > 90%; where η is the capture efficiency.

9. A computer device, characterized in that: The system includes a memory and a processor. The memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes the performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network, as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the performance prediction method for a jet-shielded flue gas capture system based on a genetic algorithm fused with a BP neural network as described in any one of claims 1-8.

Images