Nozzle atomization particle size prediction method based on GA-BP neural network

Abstract

The invention discloses a nozzle atomization particle size prediction method based on a GA-BP neural network. The method comprises the following steps: S1, obtaining external mixing type gas-liquid double-fluid nozzle atomization particle size test data; s2, data preprocessing; s3, establishing a GA-BP neural network atomization particle size prediction model: determining a topological structure of a BP neural network model; a weight value and a threshold value are generated randomly, and the initialized weight value and the threshold value are input into the genetic algorithm part; optimizing the initialized weight value and threshold value through a genetic algorithm; judging whether the number of iterations of the genetic algorithm part is equal to the maximum number of iterations, and outputting an optimal individual; inputting the optimal individual into a BP neural network, updating a weight and a threshold according to an expected error, and fixing the weight and the threshold; and S4, performing prediction by using the trained GA-BP neural network atomization particle size prediction model. The atomization particle size prediction method is high in precision, and the design period of the nozzle is shortened.

Description

Technical Field

[0001] The invention relates to the field of atomization jets, and in particular to a nozzle atomization particle size prediction method based on a GA-BP neural network. Background Art

[0002] With the rapid advancement of urbanization and industrialization in my country, air pollution has become increasingly serious. Atmospheric particulate matter is the main factor causing air pollution, causing great harm to the physical and mental health of the public.

[0003] Generating fine water mist through a gas-liquid dual-fluid nozzle can effectively improve the settling efficiency of atmospheric particulate matter. This technology has become an important way to control atmospheric particulate matter. Relevant research at home and abroad has confirmed that when the droplet size is similar to the dust particle size, the probability of condensation between the two is the greatest, and the efficiency of atomization dust removal is the highest.

[0004] Atomization particle size is one of the important indicators to measure the atomization performance of external mixing gas-liquid two-fluid nozzles, but atomization is a complex and random process. There are many factors that affect the atomization performance. The current research on the atomization mechanism is not perfect, so most of the research is based on experiments to provide practical empirical or semi-empirical atomization particle size prediction models. This model has a narrow application range and a large prediction error.

[0005] BP neural network is one of the most widely used neural network models in the field of atomization jet, but it still has research potential in terms of reliability and adaptability, which is mainly reflected in the following two aspects:

[0006] 1. The BP neural network corrects weights and thresholds based on the gradient descent method, which may cause the network to fall into local extreme points, resulting in low prediction accuracy and poor accuracy.

[0007] 2. The BP neural network prediction process has a long experimental time and a large workload, so the design cycle is long. Summary of the invention

[0008] In order to solve the problems existing in the prior art, the present invention provides a nozzle atomization particle size prediction method based on GA-BP neural network. The prediction results of this method are highly accurate and stable. Using this method to predict the nozzle atomization particle size can shorten the design cycle of external mixing gas-liquid two-fluid nozzles.

[0009] To this end, the present invention adopts the following technical solutions:

[0010] A nozzle atomization particle size prediction method based on GA-BP neural network includes the following steps:

[0011] S1, obtaining the atomization particle size test data of the external mixing gas-liquid two-fluid nozzle as the training set data, each set of training set data includes water flow rate, gas flow rate, liquid hole diameter and gas hole outer diameter;

[0012] S2, data preprocessing, normalizing the test data obtained in S1;

[0013] S3, establishing a nozzle atomization particle size prediction model of a GA-BP neural network, including the following steps:

[0014] S31, determine the topological structure of the BP neural network model:

[0015] Each group of training set data after S2 normalization is used as the input of the model, and the atomized particle size is used as the output of the model. The number of neurons in the input layer and the output layer of the BP neural network is determined according to the number of input and output features; the number of neurons in the hidden layer is determined: the value range of the number of neurons in the hidden layer is determined by an empirical formula; the BP neural network is debugged and corrected according to the value range of the number of neurons in the hidden layer, and the number of neurons in the hidden layer is determined; and the topological structure of the BP neural network is obtained;

[0016] S32, initializing the weights and thresholds of the BP neural network, and inputting the initialized weights and thresholds into the genetic algorithm part;

[0017] S33, optimizing the initialized weights and thresholds by a genetic algorithm, comprising the following steps:

[0018] S331, perform population initialization: encode the initialized weights and thresholds to obtain individuals; obtain the encoding length of the individuals based on the BP neural network topology determined in S31; determine the population size, maximum number of iterations, crossover probability and mutation probability; generate a population according to the population size;

[0019] S332, calculate the population fitness value:

[0020] The training set data after S2 preprocessing is input into the BP neural network model for training to obtain the expected output and predicted output, and then obtain the fitness value F:

[0021]

[0022] In the formula, A is the coefficient; y k and k are the expected output and predicted output of the kth group of training data of the BP neural network respectively; h1 is the number of groups of training set data;

[0023] S333, selection operation: according to the selection probability P i The individuals in the population are selected to obtain the selected population; the selection probability Pi The calculation formula is as follows:

[0024]

[0025] In the formula, B is a coefficient, in this embodiment, B=1; N is the number of individuals in the population; F i is the fitness value of individual i, calculated by formula (3) in S332, where i = 1, 2, ... N;

[0026] S334, performing a crossover operation on the selected population:

[0027] Among the individuals in the population, the kth individual a k and the lth individual a l The gene crossover operation at position j is as follows:

[0028]

[0029] Where b is a random number, and b∈[0,1]; a k j is the gene of the kth individual at position j; a lj is the gene of the lth individual at position j; is the gene at position j of the kth individual after the crossover operation; is the gene at position j of the lth individual after the crossover operation; j = 1, 2, ... 61, the value range of k and l is 1, 2, ... N;

[0030] S335, perform mutation operation on the population after the crossover operation:

[0031] Among the individuals in the population, the jth gene a of the i-th individual ij The mutation operation method is as follows:

[0032]

[0033] In the formula, a max Gene a ij The upper bound of min Gene a ij The lower bound of ; r is a random number; the calculation formula of the iterative function f(g) is as follows:

[0034] f(g)=r1(1-g / G max ) 2 (8)

[0035] In the formula, r1 is a random number; g is the current iteration number; G max is the maximum number of iterations;

[0036] S34, based on the population obtained by the mutation operation, determine whether the number of iterations at this time is equal to the maximum number of iterations: if it is equal, calculate the fitness value of each individual in the current population by formula (3) in S332, and find the individual with the lowest fitness in the current population as the optimal individual, and execute S35; if it is not equal, execute S333;

[0037] S35, inputting all the optimal weights and optimal thresholds of the optimal individuals into the BP neural network, inputting the training set data into the BP neural network for training and outputting the prediction results, calculating the expected error of the BP neural network prediction results, and continuously updating the weights and thresholds according to the expected error until the expected error of the BP neural network is less than the set value, fixing the weights and thresholds at this time, and completing the construction of the nozzle atomization particle size prediction model of the GA-BP neural network;

[0038] S4, using the model obtained in S35 to predict the atomized particle size of the external mixing gas-liquid two-fluid nozzle.

[0039] Preferably, the topological structure of the BP neural network model determined in step S31 is 4-10-1;

[0040] Among them, the method for determining the topological structure of the BP neural network model is as follows: each group of training set data contains four input features, namely water flow, gas flow, liquid pore diameter and pore outer diameter, so the number of input layer neurons is 4; the atomized particle size is used as the output of the BP neural network model, so the number of output layer neurons is 1; the value range of the number of hidden layer neurons is determined by an empirical formula, and the value range is 3-13. In this value range, the number of hidden layer neurons corresponding to the minimum mean square error MSE and the shortest training time is selected as 10 as the number of hidden layer neurons.

[0041] In step S331, the initialized weights and thresholds are encoded by the real number method, the individual coding length is 61, and the coding length q is determined by the formula: q=u×v+v+v×w+w, wherein the number of neurons in the input layer is u, the number of neurons in the hidden layer is v, and the number of neurons in the output layer is w; the population size is 10, the maximum number of iterations is 100, the crossover probability is 0.4, and the mutation probability is 0.1.

[0042] Preferably, in step S332, coefficient A=1, h1=104.

[0043] Preferably, in step S333, coefficient B=1; and the roulette method is used for the selection operation.

[0044] Preferably, in step S335, r∈[0,1].

[0045] Preferably, the setting value in step S35 is 1×10 -5 , the training rate of the BP neural network is 0.01.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. The present invention can better reflect the nonlinear relationship between various parameters and atomized particle size. Compared with the empirical or semi-empirical atomized particle size prediction model obtained through a large amount of experimental data, the accuracy of the atomized particle size prediction model of the present invention has been significantly improved, which provides an important reference for the optimization design of the external mixing gas-liquid two-fluid nozzle;

[0048] 2. The nozzle atomization particle size prediction method of the present invention effectively shortens the nozzle design cycle. Compared with the traditional experimental design, the nozzle atomization particle size prediction method of the present invention shortens the experimental time and workload to 1 / 16, greatly reduces the nozzle optimization design time, shortens the entire nozzle design cycle, and has great practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart of a nozzle atomization particle size prediction model using a GA-BP neural network established in the present invention;

[0050] Figure 2 is the fitness value of the optimal individual in the embodiment of the present invention;

[0051] Figure 3 It is a comparison diagram of the prediction results of the nozzle atomization particle size of the GA-BP neural network in the embodiment of the present invention and the nozzle atomization particle size of the existing BP neural network;

[0052] Figure 4 This is the error analysis diagram of the nozzle atomization particle size prediction method of the GA-BP neural network and the existing BP neural network. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments.

[0054] Example

[0055] A nozzle atomization particle size prediction method based on GA-BP neural network includes the following steps:

[0056] S1, obtain the test data of atomization particle size of external mixing gas-liquid two-fluid nozzle.

[0057] The data set obtained in this embodiment includes 124 groups of test data, each group of test data includes: water flow rate, gas flow rate, liquid pore diameter and pore outer diameter, of which there are 104 groups of training set data and 20 groups of test set data.

[0058] S2, data preprocessing, normalizes the test data obtained in S1 to avoid the influence of different value ranges and units of the test data on the training speed and calculation accuracy of the GA-BP neural grid. The formula for normalization is as follows:

[0059]

[0060] In the formula, x′ is the normalized data; x is the test data; x max is the maximum value in the test data; x min is the minimum value in the original data.

[0061] S3, establish a nozzle atomization particle size prediction model based on GA-BP neural network, see Figure 1 , the specific steps are as follows:

[0062] S31, determining the topological structure of the BP neural network model.

[0063] The BP neural network is usually composed of an input layer, a hidden layer, and an output layer, wherein each set of training set data (water flow, gas flow, liquid pore diameter, and pore outer diameter) after S2 normalization is used as the input of the model, and the atomized particle size is used as the output of the model. The structure of the BP neural network is determined according to the number of input and output features. Therefore, in this embodiment, the number of neurons in the input layer is 4, and the number of neurons in the output layer is 1.

[0064] The number of neurons in the hidden layer v is determined by the following empirical formula:

[0065]

[0066] Among them, a is a constant with a value range of 1-10; u is the number of neurons in the input layer, w is the number of neurons in the output layer, and v is rounded up to 3-13.

[0067] The BP neural network is debugged and corrected according to the value range of the number of hidden layer neurons determined by the above empirical formula. In this embodiment, the number of hidden layer neurons corresponding to the minimum mean square error MSE and the shortest training time is selected as the final number of hidden layer neurons, so the topological structure of the BP neural network is 4-10-1.

[0068] S32, after determining the topological structure of the BP neural network, randomly generate weights and thresholds, complete the initialization process, and input the initialized weights and thresholds into the genetic algorithm part.

[0069] S33, optimizing the initialized weights and thresholds by genetic algorithm: firstly, the initialized weights and thresholds are initialized, the genetic algorithm solves the individual fitness value through the fitness function, and then finds the individual with the best fitness value through selection operation, crossover operation and mutation operation. Specifically, the following steps are included:

[0070] S331, population initialization. In this embodiment, the population size is set to 10, the maximum number of iterations is 100, the crossover probability is 0.4, and the mutation probability is 0.1. The encoding method of the initialized thresholds and weights is: the chromosome individuals are encoded by the real number method, and each chromosome individual is a real number string containing all weights and thresholds of the BP neural network. The encoding length of the real number string is determined according to the BP neural network structure.

[0071] In this embodiment, the BP neural network topology structure determined based on S31 is 4-10-1, where the number of input layer neurons is u, the number of hidden layer neurons is v, the number of output layer neurons is w, and the coding length of the genetic algorithm individual is q=u×v+v+v×w+w. Substituting it into the formula for calculation, it can be determined that the coding length of the individuals in the genetic algorithm population is 61.

[0072] S332, calculate the population fitness value:

[0073] The training set data after S2 preprocessing is input into the BP neural network model for training, and the expected output and predicted output are obtained. The absolute value of the prediction error is calculated, and the sum of the absolute values ​​of the prediction errors of the training set data is used as the fitness value F of the individuals in the population. The calculation formula is as follows:

[0074]

[0075] Wherein, A is a coefficient, and in this embodiment, A=1; y k and k are respectively the expected output and predicted output of the kth group of training data of the BP neural network; h1 is the number of groups of training set data. In this embodiment, h1=104.

[0076] S333, select operation:

[0077] In this embodiment, the roulette method is used for selection operation, and the selection probability P i As shown below:

[0078]

[0079] In the formula, B is a coefficient, in this embodiment, B=1; N is the number of individuals in the population; F iis the fitness value of individual i, calculated by formula (3) in S332, where i = 1, 2, ... N.

[0080] According to the selection probability P i , the roulette method is used to select individuals in the population to obtain the selected population.

[0081] S334, performing a crossover operation on the selected population:

[0082] In the population of this embodiment, the kth individual a k and the lth individual a l The gene crossover operation at position j is as follows:

[0083]

[0084] Where b is a random number, and b∈[0,1]; a k j is the gene of the kth individual at position j; a lj is the gene of the lth individual at position j; is the gene at position j of the kth individual after the crossover operation; is the gene at position j of the lth individual after the crossover operation; j = 1, 2, ... 61, the value range of k and l is 1, 2, ... N.

[0085] S335, performing a mutation operation on the population after the crossover operation:

[0086] In the population of individuals in this embodiment, the jth gene a of the i-th individual ij The mutation operation method is as follows:

[0087]

[0088] In the formula, a max Gene a ij The upper bound of min Gene a ij The lower bound of ; r is a random number, and r∈[0,1]; the calculation formula of the iterative function f(g) is as follows:

[0089] f(g)=r1(1-g / G max ) 2 (8)

[0090] In the formula, r1 is a random number; g is the current iteration number; G max is the maximum number of iterations, which is 100 in this embodiment.

[0091] S34, based on the population obtained by the mutation operation, determine whether the number of iterations at this time is equal to the maximum number of iterations. If it is equal, calculate the fitness value of each individual in the current population by formula (3) in S332, and find the individual with the lowest fitness in the current population as the optimal individual, and execute S35; if it is not equal, execute S333. Figure 2 As shown, in this embodiment, the fitness value of the optimal individual is 0.27.

[0092] S35, inputting all the optimal weights and optimal thresholds of the optimal individual into the BP neural network, inputting the training set data into the BP neural network for training and outputting the prediction results, calculating the expected error of the BP neural network prediction results, and continuously updating the weights and thresholds according to the expected error until the expected error of the BP neural network is less than 1×10 -5 , that is, the BP neural network converges. The weights and thresholds at this time are fixed to complete the construction of the GA-BP neural network nozzle atomization particle size prediction model.

[0093] The BP neural network training rate is selected as 0.01. If the selected training rate is too large, the adjustment range of the weights and thresholds will be too large, reducing the stability of the BP neural network; if the selected training rate is too small, the adjustment range of the weights and thresholds will be too small, reducing the training efficiency of the BP neural network.

[0094] The selection of the expected error needs to be combined with the specific BP neural network structure and the actual training efficiency. In this embodiment, the expected error of the BP neural network is set to 1×10 -5 .

[0095] In order to comprehensively evaluate the overall accuracy of the nozzle atomization particle size prediction model based on the GA-BP neural network, the test set data is used as the input of the GA-BP neural network nozzle atomization particle size prediction model, and the prediction results are output through the GA-BP neural network to evaluate the prediction model.

[0096] The following evaluation indicators are used to evaluate the prediction results:

[0097] 1. Use the coefficient of determination R 2 To evaluate the fit between the predicted value and the experimental value, R 2 ∈[0,1],R 2 The closer it is to 1, the better the fit between the predicted value and the experimental value of the prediction model. 2 The calculation formula is as follows:

[0098]

[0099] Wherein, h2 is the number of test set data groups. In this embodiment, h2=20; yk , and k They represent the expected output, average expected output and predicted output of the kth group of test data of the BP neural network respectively.

[0100] 2. Use mean absolute error (MAE), mean square error (MSE) and mean absolute percentage error (MAPE) to evaluate the performance of the prediction model. The smaller the MSE, the closer the MAE and MAPE are to 0, indicating that the relative error of the prediction model is smaller and the prediction accuracy is higher. The calculation formulas of MAE, MSE and MAPE are as follows:

[0101]

[0102] Where, h2 is the number of test set data groups. In this embodiment, h2 = 20; y k and k They represent the expected output and predicted output of the kth group of test data of the BP neural network respectively.

[0103] In this embodiment, the evaluation indicators of the nozzle atomization particle size prediction results based on the GA-BP neural network are shown in Table 1.

[0104] Table 1 Evaluation of nozzle atomization particle size prediction results

[0105]

[0106] It can be seen from the table that the determination coefficient R of the GA-BP neural network nozzle atomization particle size prediction model 2 It is 0.99, indicating that the GA-BP neural network has a good fitting degree and a high prediction accuracy. The MAE, MSE and MAPE of the nozzle atomization particle size prediction model of the GA-BP neural network are 1.091, 3.32 and 0.018, respectively, indicating that the error of the prediction result of the nozzle atomization particle size prediction model of the GA-BP neural network is small.

[0107] It can be seen that the prediction results of the nozzle atomization particle size prediction model proposed in the present invention have high accuracy and small error, and can accurately reflect the interaction between the various input parameters of the prediction model and the atomization particle size, which provides an important reference for the optimization design of the external mixing gas-liquid two-fluid nozzle.

[0108] Compare the prediction results of the GA-BP neural network nozzle atomization particle size prediction model with the BP neural network nozzle atomization particle size prediction model:

[0109] The comparison between the nozzle atomization particle size prediction results of GA-BP neural network and BP neural network and the test results is shown in Figure 2. Figure 3 As shown, Figure 3The vertical coordinate SMD is the Sauter mean particle size, which indicates the particle size. Compared with the nozzle atomization particle size prediction value of the BP neural network, the nozzle atomization particle size prediction value of the GA-BP neural network has a better fit with the experimental value (basically coincident), and can more accurately predict the atomization particle size of the external mixing gas-liquid dual-fluid nozzle.

[0110] The error of the nozzle atomization particle size prediction results of the GA-BP neural network is compared and analyzed, such as Figure 4 As shown, compared with the existing BP neural network, the relative error of the prediction value of the GA-BP neural network of the present invention is reduced from ±14% to ±7%, and the improvement effect is very significant.

[0111] It can be seen that optimizing the initial weights and thresholds of the BP neural network through genetic algorithms can effectively improve the accuracy of the nozzle atomization particle size prediction model.

[0112] S4, using the trained GA-BP neural network nozzle atomization particle size prediction model to make predictions: the water flow rate, gas flow rate, liquid hole diameter and air hole outer diameter are input into the trained GA-BP neural network nozzle atomization particle size prediction model to predict the atomization particle size, thereby completing the design of the external mixing gas-liquid two-fluid nozzle.

Claims

1. A nozzle atomization particle size prediction method based on GA-BP neural network, characterized in that The following steps are involved: S1, obtaining the atomization particle size test data of the external mixing gas-liquid two-fluid nozzle as the training set data, each set of training set data includes water flow rate, gas flow rate, liquid hole diameter and gas hole outer diameter; S2, data preprocessing, normalizing the test data obtained in S1; S3, establishing a nozzle atomization particle size prediction model of a GA-BP neural network, including the following steps: S31, determine the topological structure of the BP neural network model: Each group of training set data after S2 normalization is used as the input of the model, and the atomized particle size is used as the output of the model. The number of neurons in the input layer and the output layer of the BP neural network is determined according to the number of input and output features; the value range of the number of neurons in the hidden layer is determined by an empirical formula; the BP neural network is debugged and corrected according to the value range of the number of neurons in the hidden layer to determine the number of neurons in the hidden layer; and the topological structure of the BP neural network is obtained; S32, initializing the weights and thresholds of the BP neural network, and inputting the initialized weights and thresholds into the genetic algorithm part; S33, optimizing the initialized weights and thresholds by a genetic algorithm, comprising the following steps: S331, perform population initialization: encode the initialized weights and thresholds to obtain chromosome individuals; obtain the encoding length of the chromosome individuals based on the BP neural network topology determined in S31; determine the population size, maximum number of iterations, crossover probability and mutation probability; generate a population according to the population size; S332, calculate the population fitness value: The training set data after S2 preprocessing is input into the BP neural network model for training to obtain the expected output and predicted output, and then obtain the fitness value F: In the formula, A is the coefficient; y k and k are the expected output and predicted output of the kth group of training data of the BP neural network respectively; h1 is the number of groups of training set data; S333, selection operation: according to the selection probability P i The chromosome individuals in the population are selected to obtain the selected population; the selection probability P i The calculation formula is as follows: In the formula, B is a coefficient, in this embodiment, B=1; N is the number of chromosome individuals in the population; F i is the fitness value of chromosome individual i, calculated by formula (3) in S332, where i = 1, 2, ... N; S334, performing a crossover operation on the selected population: Among the individuals in the population, the kth chromosome individual a k and chromosome individual a l The gene crossover operation at position j is as follows: Where b is a random number, and b∈[0,1]; a kj is the gene at position j of the kth chromosome individual; a lj is the gene at position j of the lth chromosome individual; is the gene at position j of the kth chromosome individual after the crossover operation; is the gene at position j of the individual on the lth chromosome after the crossover operation; j = 1, 2, ... 61, the value range of k and l is 1, 2, ... N; S335, perform mutation operation on the population after the crossover operation: Among the chromosome individuals in the population, the jth gene a of the i-th chromosome individual ij The mutation operation method is as follows: In the formula, a max Gene a ij The upper bound of min Gene a ij The lower bound of ; r is a random number; the calculation formula of the iterative function f(g) is as follows: f(g)=r1(1-g / G max ) 2 (8) In the formula, r1 is a random number; g is the current iteration number; G max is the maximum number of iterations; S34, based on the population obtained by the mutation operation, determine whether the number of iterations at this time is equal to the maximum number of iterations: if it is equal, calculate the fitness value of each chromosome individual in the current population by formula (3) in S332, and find the chromosome individual with the lowest fitness in the current population as the optimal individual, and execute S35; if it is not equal, execute S333; S35, inputting all the optimal weights and optimal thresholds of the optimal individuals into the BP neural network, inputting the training set data into the BP neural network for training and outputting the prediction results, calculating the expected error of the BP neural network prediction results, and continuously updating the weights and thresholds according to the expected error until the expected error of the BP neural network is less than the set value, fixing the weights and thresholds at this time, and completing the construction of the nozzle atomization particle size prediction model of the GA-BP neural network; S4, using the model obtained in S35 to predict the atomized particle size of the external mixing gas-liquid two-fluid nozzle.

2. The nozzle atomization particle size prediction method according to claim 1, characterized in that: In step S31, the method for determining the topological structure of the BP neural network model is as follows: Each group of training set data contains four input features: water flow, gas flow, liquid pore diameter and pore outer diameter, so the number of input layer neurons is 4; the atomized particle size is the output of the BP neural network model, so the number of output layer neurons is 1; the value range of the number of hidden layer neurons is determined by an empirical formula, and the value range is 3-13. In this value range, the number of hidden layer neurons corresponding to the minimum mean square error MSE and the shortest training time is selected as the number of hidden layer neurons.

3. The nozzle atomization particle size prediction method according to claim 1, characterized in that: In step S331, the initialized weights and thresholds are encoded by the real number method, the coding length of the chromosome individual is 61, and the coding length q is determined by the formula: q=u×v+v+v×w+w, where the number of neurons in the input layer is u, the number of neurons in the hidden layer is v, and the number of neurons in the output layer is w.

4. The nozzle atomization particle size prediction method according to claim 1, characterized in that: In step S331, the population size is 10, the maximum number of iterations is 100, the crossover probability is 0.4, and the mutation probability is 0.

1.

5. The nozzle atomization particle size prediction method according to claim 1, characterized in that: In step S332, A=1; h1=104.

6. The nozzle atomization particle size prediction method according to claim 1, characterized in that: In step S333, B=1; the roulette method is used for selection operation.

7. The nozzle atomization particle size prediction method according to claim 1, characterized in that: In step S335, r∈[0,1].

8. The nozzle atomization particle size prediction method according to claim 1, characterized in that: In step S35, the setting value is 1×10 -5 , the training rate of the BP neural network is 0.01.

Images