Multi-objective intelligent optimization method for biomass supercritical water fluidized bed reactor

Abstract

The invention relates to a multi-target intelligent optimization method of a biomass supercritical water fluidized bed reactor. A two-fluid model is coupled with a chemical reaction model to carry out analog simulation on a reaction process of supercritical water and biomass under different operation parameters and reactor structure parameters, and the gas yield of gases such as hydrogen and the like is obtained; dividing data obtained by simulation into a training set, a test set and a verification set; constructing a neural network prediction model by adopting the training set, and evaluating the generalization ability of the ANN-GA neural network prediction model by adopting cross validation; and testing the precision of the neural network prediction model by using the test set and performing index evaluation. According to the method, the predicted value of the gas yield of the biomass supercritical water fluidized bed can be obtained according to industrial requirements, the process parameters and the structure combination under the expected target can be reversely obtained, the optimization time can be greatly saved, and the optimization cost of the biomass supercritical water fluidized bed reactor can be reduced.

Description

Technical Field

[0001] The present invention relates to the fields of structural optimization, simulation optimization, and chemical engineering of biomass supercritical water fluidized bed reactors, and particularly relates to a multi-objective intelligent optimization method for biomass supercritical water fluidized bed reactors. Background Art

[0002] The energy consumption of traditional blast furnace ironmaking process accounts for more than 70% of the total steel production process. A large amount of atmospheric pollutants such as CO2, SO2, and NO are generated in the traditional iron and steel metallurgy process. The low-carbon transformation of iron and steel metallurgy is the trend of the development of metallurgical technology. Hydrogen metallurgy technology is one of the important measures to achieve carbon neutrality in the iron and steel industry. China lacks the hydrogen source required for direct reduced iron. To solve the problems faced by China, it is necessary to find a low-carbon, inexpensive, and abundantly available hydrogen resource. x The supercritical water hydrogen production technology can convert the rich biomass resources in China into hydrogen without pollution and with high efficiency. The most widely used reactor in the supercritical water hydrogen production technology is the fluidized bed. The advantage of the fluidized bed is that it enables the reactants to be fully mixed, improves the reaction efficiency, and prevents the reactor from being blocked. It can achieve zero emissions of nitrogen hydrides, sulfides, solid particles, and related pollutants from the source, and is a green and potential hydrogen production technology.

[0003] Experimental research on the optimization of supercritical water fluidized bed reactors requires building experimental platforms for supercritical water fluidized bed reactors of multiple different scales, which is costly and limited by measurement means and cannot be fully characterized. Therefore, the structural optimization of supercritical water fluidized bed reactors relies too much on trial and error, resulting in low optimization efficiency and high optimization costs. Therefore, numerical simulation and machine learning have been introduced into the field of structural optimization of fluidized beds.

[0004]

[0005] ​CN118230838A discloses a multi-objective adaptive intelligent optimization algorithm integrating the mechanism of a methanol production system, which includes the following steps: (1) Using existing simulation tools to establish a mechanism model corresponding to a complete set of methanol production systems including triple carbon dioxide feed, steam methane reforming (SMR) reactor, dry methane reforming (DMR) reactor, methanol synthesis reactor, and two distillation columns; (2) Selecting n key process variables as decision variables, and using the Latin hypercube sampling (LHS) method to generate a decision variable data set corresponding to the decision variables. The data in the decision variable data set constitutes the input of the mechanism model corresponding to the methanol production system, and the decision variable data set contains M sampling points; (3) Inputting the decision variable data set into the established mechanism model and executing it to obtain the output variables corresponding to the methanol production system; (4) Normalizing the data; (5) Using a convolutional neural network (CNN) to learn the internal laws of the data normalized in step (4) and establishing a surrogate model in the deep learning hybrid framework; (6) Using the reference point-based non-dominated sorting genetic algorithm (NSGA-III) and combining with the surrogate model in step (5) to complete the multi-objective optimization of four performance indicators of the methanol production system process; (7) Using the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) to determine the optimal solution. By deeply coupling the surrogate model and the meta-heuristic algorithm, it is used to predict the behavior of complex chemical systems and optimize the performance of the system. However, the convolutional neural network algorithm relies on a large amount of data and computing resources, and is prone to overfitting when the data is insufficient, resulting in poor model accuracy, and there are limitations and defects.

[0006] The present invention uses a two-fluid model to couple a chemical model to simulate chemical reactions. On this basis, the genetic neural network (ANN-GA) technology is used to perform multi-objective intelligent optimization on a biomass supercritical water fluidized bed reactor, which is expected to solve the problems of low optimization efficiency and high optimization cost of the biomass supercritical water fluidized bed reactor. Summary of the Invention

[0007] To solve the above technical problems, the present invention provides a method for multi-objective intelligent optimization of a biomass supercritical water fluidized bed reactor. The method provided by the present invention can obtain the predicted value of the gas production of the biomass supercritical water fluidized bed according to industrial needs, and inversely obtain the process parameter and structure combination under the desired target through the prediction model. Compared with the traditional manual optimization, it has the advantages of simple operation, high efficiency, and high accuracy.

[0008] To achieve this purpose, the present invention adopts the following technical solutions:

[0009] The present invention provides a method for multi-objective intelligent optimization of a biomass supercritical water fluidized bed reactor, and the method includes the following steps:

[0010] (1) Determine the structural parameters of the biomass supercritical water fluidized bed reactor;

[0011] (2) Set the operating parameters of the biomass supercritical water fluidized bed reactor and the physical property parameters of the fluid in the simulation software, and obtain the output parameters through simulation calculations according to the structural parameters and operating parameters.

[0012] (3) Collect data and establish a dataset of the structural parameters and operating parameters of the biomass supercritical water fluidized bed reactor, normalize the data in the dataset, and divide the data in the dataset.

[0013] (4) Construct a neural network prediction model through the dataset described in step (3).

[0014] (5) Verify the accuracy of the neural network prediction model described in step (4) and evaluate it with evaluation indicators; after the accuracy meets the standard, introduce the GA genetic algorithm and combine it with the ANN neural network to output the ANN-GA genetic neural network prediction model.

[0015] (6) Optimize the biomass supercritical water fluidized bed reactor according to the genetic neural network prediction model described in step (5).

[0016] The method of the present invention uses the genetic neural network (ANN-GA) technology to establish a neural network model based on feedforward backpropagation. The neural network uses a three-layer BP network for modeling, determines the number of nodes in the hidden layer using a formula, clarifies the relationship and weight of various structural and operating parameters on the gas production and bed pressure drop, combines the established neural network model with the genetic algorithm to optimize the efficiency and accuracy of the model, and obtains the prediction model ANN-GA of the relationship between the operating parameters and structure of the fluidized bed and the gas production, bed pressure drop, and total calorific value of supercritical water. The method of the present invention can save a large amount of optimization time and reduce the optimization cost of the biomass supercritical water fluidized bed reactor.

[0017] As a preferred technical solution of the present invention, the structural parameters of the supercritical water fluidized bed reactor described in step (1) include: the angle of the air distribution plate, the initial bed material height, the angle of the feed inlet, and the initial solid volume fraction.

[0018] It should be noted that the structural parameters of the supercritical water fluidized bed reactor selected in the present invention are the core parameters for model calculation, but are not limited to the listed parameters, and other possible parameters are also included.

[0019] As a preferred technical solution of the present invention, the operating parameters of the biomass supercritical water fluidized bed reactor described in step (2) include: the supercritical water inlet velocity, the raw material inlet velocity, and the wall temperature.

[0020] Preferably, the physical property parameters of the fluid include density, viscosity, and specific heat capacity.

[0021] Preferably, the output parameters in step (2) include: gas production, bed pressure drop, and total calorific value of supercritical water.

[0022] As a preferred technical solution of the present invention, the simulation calculation in step (2) includes the following steps:

[0023] (1’) Modeling of the biomass supercritical water fluidized bed reactor is a simplified two-dimensional model, and the modules are optimized;

[0024] (2’) Set relevant physical property parameters, and use a chemical reaction model to simulate the chemical reactions in the biomass supercritical water fluidized bed reactor; when setting boundary conditions, set initial conditions;

[0025] (3’) In the post-processing stage, calculate the output parameters.

[0026] Preferably, the optimization in step (1’) includes the optimization of redundant modules with less influence on the results.

[0027] Preferably, the relevant physical property parameters in step (2’) include: biomass, supercritical water, and gas materials, and the diffusion coefficient, viscosity, density, and specific heat parameters of the materials in the mixed phase are fitted in combination with experimental data.

[0028] Preferably, the initial conditions in step (2’) include: inlet, outlet, wall thickness, and temperature.

[0029] As a preferred technical solution of the present invention, the data set in step (3) includes: the initial bed height, initial solid volume fraction, feed inlet angle, and air distribution plate inclination angle of the biomass supercritical water fluidized bed reactor. Values are taken for each parameter and arranged in combinations to obtain data groups.

[0030] Preferably, three values are selected for each parameter and arranged in combinations to calculate a total of 81 groups of data, namely 3×3×3×3, to obtain data groups.

[0031] Preferably, the normalization formula in step (3) is:

[0032]

[0033] where y i is the normalized data, y is the original data, y max and y min are the maximum and minimum values of the data.

[0034] Preferably, the division of the data in the data set in step (3) includes: dividing the data in the data set into a training set, a validation set, and a test set according to a ratio.

[0035] More preferably, the data in the data set is divided into a training set, a validation set, and a test set according to a ratio of 4:1:1.

[0036] Preferably, the training set is used to construct the neural network prediction model described in step (4).

[0037] Preferably, the validation set is used to evaluate the generalization ability of the neural network prediction model described in step (4).

[0038] Preferably, the test set is used to test the accuracy of the neural network prediction model described in step (5).

[0039] When constructing a model using the training set in the present invention, a neural network prediction model is selected for construction. Since there is a complex non-linear relationship between the input parameters (structural parameters and physical property parameters) and the output parameters, the neural network model can avoid falling into the local optimum problem and is more likely to find the global optimum solution.

[0040] As a preferred technical solution of the present invention, the evaluation index in step (5) is the coefficient of determination R 2 and / or the root mean square error RMSE.

[0041] Preferably, the calculation formulas for the coefficient of determination R 2 and the root mean square error RMSE are as follows:[[]]

[0042]

[0043] where N is the total number of samples, z i , are the experimental value, the predicted output, and the average value of the predicted output, respectively.

[0044] As a preferred technical solution of the present invention, the optimization steps in step (6) include: calculating the gas production amount and the bed pressure drop of the chemical reaction in the biomass supercritical water fluidized bed reactor according to the genetic neural network prediction model, and inferring whether the design requirements are met, so as to determine the operating parameters and structural parameters of the biomass supercritical water fluidized bed reaction that meet the requirements.

[0045] Compared with the prior art, the present invention has at least the following beneficial effects:

[0046] When optimizing the structure of the biomass supercritical water fluidized bed reactor, the method of the present invention adopts an optimization algorithm based on the gradient descent algorithm to search for the optimal structure of the fluidized bed under the given constraints, improving the uniformity of particle distribution, enhancing the optimization efficiency of the fluidized bed, reducing the optimization cost of the fluidized bed, and providing guidance for the optimization of the fluidized bed. Brief Description of the Drawings

[0047] Figure 1 It is a flow chart of an intelligent optimization method for a biomass supercritical water fluidized bed reactor.

[0048] Figure 2 It is a structural diagram of a simplified biomass supercritical water fluidized bed reactor. Specific implementation mode

[0049] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings and through specific implementation modes. However, the following examples are only simple examples of the present invention and do not represent or limit the scope of the protection of the present invention. The scope of protection of the present invention shall be subject to the claims.

[0050] Example 1

[0051] This example provides a method for multi-objective intelligent optimization of a biomass supercritical water fluidized bed reactor. The flowchart of the method is as Figure 1 shown, and the method includes the following steps:

[0052] (1) Use modeling software to construct a two-dimensional simplified model of the biomass supercritical water fluidized bed reactor, as Figure 2 shown. The operating parameters are the supercritical water inlet velocity V1, the raw material inlet velocity V2, the wall temperature T1, and the structural parameters are the wall thickness H1 and the wall temperature T1. Among them, 0.02 m / s < V1 < 0.06 m / s, 0.0067 m / s < V2 < 0.02 m / s, 773 K < T1 < 873 K, 0.0105 m < H1 < 0.0505 m. Subsequently, different orthogonal combinations are generated, and the data of different operating parameters and structural parameters are calculated. The optimization parameters and optimization objectives are shown in Table 1;

[0053] (2) Import the model into the mesh generation software and simulation software, perform mesh generation and assign the properties of the two-phase fluid materials in the fluidized bed, and set the boundary conditions;

[0054] (3) Simulate the hydrogen production process of the reaction between supercritical water and biomass in the fluidized bed. Calculate according to different operating parameters and structural parameters respectively to obtain the gas production and bed pressure drop after each model simulation, and construct a dataset of the operating parameters and structural parameters of the biomass supercritical water fluidized bed reactor. Perform preliminary screening and normalization on the data, normalize the data according to the following formula, and divide it into a training set, a validation set, and a test set according to the ratio of 4:1:1;

[0055] The normalization formula is:

[0056]

[0057] where y i is the normalized data, y is the original data, y max and y min are the maximum and minimum values of the data;

[0058] (4) A machine learning prediction model is constructed using the training set. The operating parameters and structural parameters (V1, V2, T1, H1) of the biomass supercritical water fluidized bed reactor are used as input parameters, and the gas production, bed pressure drop, and total calorific value of supercritical water are calculated through the ANN-GA genetic neural network prediction model. The accuracy of the neural network prediction model is evaluated using evaluation indicators, and the generalization ability of the neural network prediction model is evaluated using cross-validation;

[0059] The evaluation indicators are the coefficient of determination R 2 and / or the root mean square error RMSE. The calculation formulas for the coefficient of determination R 2 and the root mean square error RMSE are as follows:

[0060]

[0061] where N is the total number of samples, z i , are the experimental value, predicted output, and average value of the predicted output, respectively;

[0062] (5) The test set is used to test and verify the accuracy of the machine learning prediction model and evaluate it using evaluation indicators. When the accuracy meets the standard, the neural network prediction model is output. When the accuracy does not meet the standard, step (5) is repeated until the accuracy meets the standard; after the accuracy meets the standard, the ANN-GA genetic neural network prediction model is output, which is the prediction model for the gas production and bed pressure drop of the biomass supercritical water fluidized bed reactor;

[0063] (6) Based on the neural network prediction of the gas production and bed pressure drop of the chemical reaction in the biomass supercritical water fluidized bed reactor, it is deduced whether the design requirements are met, so as to determine the operating parameters and structural parameters that meet the biomass supercritical water fluidized bed reaction.

[0064] Table 1

[0065] X (Operating parameters and structural parameters) Y (Optimization objective) Initial solid volume fraction Gas production Air distributor angle Bed pressure drop Feed inlet angle Total calorific value of supercritical water Initial bed height

[0066] In summary, the present invention provides a method for multi-objective intelligent optimization of a biomass supercritical water fluidized bed reactor. The structural parameter and operating parameter datasets of the biomass supercritical water fluidized bed reactor are constructed using numerical simulation. The data in the datasets are preprocessed, and after data normalization, they are divided into a training set, a validation set, and a test set according to a certain ratio. The ANN-GA genetic neural network is used to construct a prediction model for the structural parameters and operating parameters of the biomass supercritical water fluidized bed reactor. The generalization ability of the model is evaluated through the validation set to prevent overfitting, and then the test set is used to test the reliability of the model. The correlation coefficient R 2, numerical indicators such as the root mean square error (RMSE) are used to evaluate the accuracy of the model, and finally the best prediction model is obtained. The prediction model in the present invention can predict the gas production of the biomass supercritical water fluidized bed reactor, so as to obtain the structural parameters and operating parameters of the biomass supercritical water fluidized bed reactor, thereby improving the optimization efficiency of the biomass supercritical water fluidized bed reactor.

[0067] The applicant declares that the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any changes or substitutions that can be easily thought of by any person skilled in the art within the technical scope disclosed by the present invention fall within the protection scope and the disclosure scope of the present invention.

Claims

1. A multi-objective intelligent optimization method for a biomass supercritical water fluidized bed reactor, characterized in that: The method comprises the following steps: (1) Determine the structural parameters of the biomass supercritical water fluidized bed reactor; (2) setting the operating parameters of the biomass supercritical water fluidized bed reactor and the physical parameters of the fluid in the simulation software, and obtaining the output parameters through simulation calculation according to the structural parameters and the operating parameters; (3) collecting data and establishing a data set of structural parameters and operating parameters of a biomass supercritical water fluidized bed reactor, normalizing the data in the data set, and dividing the data in the data set; (4) constructing a neural network prediction model using the data set described in step (3); (5) verifying the accuracy of the neural network prediction model in step (4) and evaluating it with evaluation indicators; after the accuracy reaches the standard, introducing the GA genetic algorithm and combining it with the ANN neural network to output the ANN-GA genetic neural network prediction model; (6) Optimizing the biomass supercritical water fluidized bed reactor according to the genetic neural network prediction model described in step (5).

2. The method according to claim 1, characterized in that: The structural parameters of the supercritical water fluidized bed reactor in step (1) include: the angle of the air distribution plate, the initial bed material height, the angle of the feed port, and the initial solid volume fraction.

3. The method according to claim 1 or 2, characterized in that: The operating parameters of the biomass supercritical water fluidized bed reactor in step (2) include: supercritical water inlet velocity, raw material inlet velocity and wall temperature; The physical properties of the fluid include density, viscosity and specific heat capacity.

4. The method according to any one of claims 1 to 3, characterized in that: The output parameters of step (2) include: gas production, bed pressure drop and total calorific value of supercritical water.

5. The method according to any one of claims 1 to 4, characterized in that: The simulation calculation in step (2) includes the following steps: (1') The modeling of the biomass supercritical water fluidized bed reactor is a simplified two-dimensional model, and the modules are optimized; (2') Setting relevant physical property parameters and using a chemical reaction model to simulate the chemical reaction in a biomass supercritical water fluidized bed reactor; when setting boundary conditions, setting initial conditions; (3') Post-processing stage, calculating output parameters; Preferably, the relevant physical property parameters in step (2') include: biomass, supercritical water and gas materials, and the diffusion coefficient, viscosity, density and specific heat parameters of the materials in the mixed phase are fitted in combination with experimental data; Preferably, the initial conditions in step (2') include: inlet, outlet, wall thickness and temperature.

6. The method according to any one of claims 1 to 5, characterized in that: The data set in step (3) includes: the initial bed height, initial solid volume fraction, feed port angle and air distribution plate inclination angle of the biomass supercritical water fluidized bed reactor. Each parameter is taken and arranged and combined to obtain a data set.

7. The method according to any one of claims 1 to 6, characterized in that: Step (3) of dividing the data in the data set includes: dividing the data in the data set into a training set, a validation set and a test set in proportion.

8. The method according to claim 7, characterized in that The training set is used to construct the neural network prediction model described in step (4); Preferably, the validation set is used to evaluate the generalization ability of the neural network prediction model described in step (4); Preferably, the test set is used to test the accuracy of the neural network prediction model described in step (5).

9. The method according to any one of claims 1 to 8, characterized in that: The evaluation index in step (5) is the determination coefficient R 2 and / or root mean square error RMSE.

10. The method according to any one of claims 1 to 9, characterized in that: The optimization step in step (6) includes: calculating the gas production and bed pressure drop of the chemical reaction of the biomass supercritical water fluidized bed reactor according to the genetic neural network prediction model, and deducing whether the design requirements are met, thereby determining the operating parameters and structural parameters of the biomass supercritical water fluidized bed reaction that meet the requirements.

Images