Main nozzle optimization method based on hybrid network and multi-strategy particle swarm optimization

Abstract

The invention discloses a main nozzle optimization method based on a hybrid network and a multi-strategy particle swarm algorithm, and belongs to the technical field of gas dynamics data processing. According to the main nozzle optimization method based on the hybrid network and the multi-strategy particle swarm algorithm, a deep learning neural network is combined with a machine learning algorithm to construct a regression model of main nozzle structure parameters and speed; an improved Newton-Raphson (INRBO) optimization algorithm is adopted to carry out hyper-parameter optimization on a convolutional neural network-attention mechanism-random forest (CNN-Attention-RF) regression model, and the speed of jet flow of a main nozzle along different monitoring points of the axis is predicted through an improved hybrid model. And then, a heterogeneous improved dynamic particle swarm (HIDMS-PSO) optimization algorithm is used for finding out main nozzle structure parameters corresponding to the highest speed predicted by the proxy model, so that the weft insertion efficiency of the air jet loom is improved, the research and development cost is reduced, the research and development period is shortened, and the method has important theoretical significance and engineering value.

Description

Technical Field

[0001] The invention relates to the technical field of airflow dynamics data processing, and in particular to a main nozzle optimization method based on a hybrid network and a multi-strategy particle swarm algorithm. Background Art

[0002] The main nozzle is a key component of air-jet weft insertion in air-jet looms. The jet characteristics of the main nozzle directly affect the efficiency, energy consumption and final quality of the fabric. In order to ensure that the air-jet loom can achieve high-speed air-jet weft insertion, the main nozzle jet needs to deliver the weft yarn to the relay area in the profiled reed at high speed and stability. The main nozzle structure design adopts a small throat section, a profiled hole guide groove and a yarn guide tube with a large aspect ratio to meet the requirements of "high speed and small flow" for weft insertion. Due to the small internal space and complex structure of the main nozzle, it is difficult to measure the internal flow field of the main nozzle using devices such as Pitot tubes and hot wire anemometers. Therefore, multi-physics computational fluid dynamics (CFD) software is used to perform integrated numerical simulation of the internal and external flow fields of the main nozzle, and the velocity characteristic curves on the centerline of the main jet and on different cross sections are obtained, revealing the influence of the main nozzle structural parameters on the jet velocity distribution. Traditional optimization methods use empirical formulas or response surface optimization methods, which are insufficient in fitting complex fluids and low in efficiency in processing high-dimensional variables, or fall into local optimality, and cannot obtain the optimal main nozzle structural parameters. Summary of the invention

[0003] The purpose of the present invention is to provide a main nozzle optimization method based on a hybrid network and a multi-strategy particle swarm algorithm. A regression model of the main nozzle mechanism parameters and speed is constructed by combining a deep learning neural network with a machine learning algorithm. The improved INRBO optimization algorithm is used to perform hyperparameter optimization on the CNN-Attention-RF regression model. The speed of the main nozzle jet at different points along the axis is predicted by the improved model, and then the HIDMS-PSO algorithm is used to find the main nozzle structural parameters corresponding to the highest speed predicted by the proxy model, so as to improve the weft insertion efficiency of the jet loom and reduce the R&D cost and cycle. It has important theoretical significance and engineering value.

[0004] To achieve the above object, the present invention provides a main nozzle optimization method based on a hybrid network and a multi-strategy particle swarm algorithm, comprising the following steps: S1. Establish a three-dimensional flow field network model of the main nozzle, select the data of the design parameters, simulate the internal and external flow field distribution laws corresponding to the three-dimensional flow field network model corresponding to different main nozzle structures, extract the central axis velocity curve, and design an experimental device to verify the central axis monitoring point velocity, obtain data samples containing the axis velocity and design parameters, and pre-process the data samples; S2. Build a CNN-Attention-RF neural network model; S3. Optimize the hyperparameters of the CNN-Attention-RF neural network model based on the INRBO algorithm; S4, INRBO-CNN-Attention-RF model predicts the airflow velocity along the axis of the main nozzle with different structures through training; S5. The optimal structural parameters of the main nozzle are obtained based on the multi-strategy particle swarm optimization algorithm.

[0005] Preferably, in S1, the acquisition of data samples is specifically as follows: according to the main nozzle design parameters, a number of different structure data are sampled using advanced Latin hypercube sampling within the corresponding value range, and specific parameters such as boundary conditions and initial conditions are set using a general solver (Fluent or Star ccm+) of computational fluid dynamics (CFD), and different main nozzle jet axis velocities are obtained through CFD numerical simulation to construct training set and test set samples for deep learning. A Pitot tube experimental device is designed to test the main jet axis velocity, and the numerical results are verified to construct a verification set for deep learning.

[0006] Preferably, in S2, in the CNN-Attention-RF neural network model, a convolutional neural network (CNN) is used to perform convolution processing on the main nozzle input structural parameter data and the output axis jet velocity, extract new data features, and enhance the correlation between the input data and the output data; an attention mechanism (Attention) is used to perform weighted fusion on the features extracted by CNN so that the network can automatically focus on information that is more useful for prediction; and a random forest regression model (RF) balances the errors of unevenly distributed data in the sample to perform regression prediction.

[0007] Preferably, in S3, the INRBO algorithm specifically includes the following steps: 1) Initialization parameters; 2) Initialize the population using chaotic sequences; 3) Adaptive inertia weight calculation, using dynamically adjusted inertia weight; 4) Apply the Newton-Raphson search strategy (NRSR) to adjust the position update of the vector according to the difference between the optimal solution and the worst solution; 5) Determine whether the random number is less than the difference factor (DF); if the random number is less than the difference factor, perform TAO update; if the random number is greater than or equal to the difference factor, perform local optimal detection. When the local optimal detection detects that the improvement of the optimal solution is less than the preset threshold, increase the stagnation counter. When the stagnation counter reaches a certain number of times, it is determined that the current search is trapped in the local optimal state, triggering chaotic perturbation. Use chaotic perturbation Logistic mapping to update the position of the solution, apply perturbation to the current position, so that the particle jumps out of the local optimal area and restores the global search capability. 6) Determine whether the iteration is terminated. If the iteration is terminated, update the convergence curve to record the best result and output the result.

[0008] Preferably, in 2), the use of chaotic sequence to initialize the population is specifically as follows: The position of the initialization vector is initialized using the Tent chaotic map. By introducing a chaotic sequence, a more random and unpredictable initialization position is generated, which improves the exploration ability of the algorithm. For each vector position , randomly generated in the interval ([0,1]): ∼Uniform(0,1); Apply the Tent map to each randomly generated value: ; The chaotic sequence Map to the specified search space range: ; in, UB is the upper bound of the search space, LB is the lower bound of the search space, d It is the parameter of Tent mapping.

[0009] Preferably, in said 3), Inertia weight formula: ; in, is the inertia weight of the current iteration, is the current iteration number, is the maximum number of iterations, is the maximum value of the inertia weight, is the minimum value of the inertia weight, α To control the steepness of the Sigmoid function, β To control the center position of the Sigmoid function.

[0010] Preferably, in 4), the NRSR search specifically includes: calculate ΔX , ; Calculate the first NRSR, , is a random number sampled from a standard normal distribution.

[0011] Propose three difference operators and improve parameters ρ , guide the population to the right direction, improve the computational efficiency of the NRBO algorithm, and propose ρ The expressions for the parameters are as follows: ; ; ; ; In this formula r 1 , r 2 ,r 3 and r 4 are distinct integers randomly selected from the population, is the best vector position , is the current position of the vector , will be continuously updated.

[0012] Update Location , ; calculate and , ; Calculate the second NRSR, ; in, ΔX is a random perturbation used to adjust the current solution, NRSR is a search rule based on random perturbations. point The position is combined by random weights. Generate new search directions, Location Updates To select different update strategies based on the value of the position, The second NRSR is to further adjust the search direction and increase the diversity and flexibility of the search.

[0013] Preferably, in S5, the multi-strategy particle swarm algorithm divides the particle swarm into several units, each unit includes a master particle and several slave particles, and the particles in the same unit interact with each other and exchange information with other units.

[0014] Preferably, the initial population of the multi-strategy particle swarm algorithm is divided into two equal isomorphic subgroups and heterogeneous subgroups. The isomorphic subgroup uses the classic PSO algorithm to update the equation, and the heterogeneous subgroup forms N p The inward moving strategy guides particles to the same unit, and the outward moving strategy uses samples from different units to guide particles.

[0015] Preferably, the inward movement strategy uses the position information obtained from all unit members to provide guidance for the particles. The master particles randomly choose any of the following three equations to update their velocity: ; ; ; in: is the main particle velocity, The main particle at time t location, For the n The particle with the smallest position similarity among the units, It is n The unit is best suited for the particle position, is the average position of all slave particles in the current unit of the master particle, ω is the inertia weight, constant c 1 and c 2 is the acceleration factor, R 1 and R 2 is a random vector ∈ [0,1]; The slave particles are moved using the following formula: ; in, is the particle velocity, p best For a particle at time t The best location found, is the particle position, For the n The main particle position of each unit; The outward movement strategy provides particles with guidance from different units in the population while maintaining the hierarchy. The master particles randomly choose any of the following three equations to update their velocity: ; ; ; in, is the main particle velocity, For the n The average position of particles in a cell, is the position of the primary particle of the randomly selected unit, is the average position of the particle's own unit members; A slave particle moves toward a random slave particle of the same type belonging to another cell using the following velocity update equation: .

[0016] The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm described in the present invention utilizes deep learning neural network combined with machine learning algorithm to construct a regression model of main nozzle mechanism parameters and speed, and adopts improved INRBO optimization algorithm to perform hyperparameter optimization on CNN-Attention-RF regression model. The speed of main nozzle jet at different points along the axis is predicted by improved model, and then HIDMS-PSO algorithm is used to find the main nozzle structure parameters corresponding to the highest speed predicted by the proxy model, so as to improve the weft insertion efficiency of air jet loom and reduce R&D cost and cycle, which has important theoretical significance and engineering value.

[0017] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic diagram of the network structure of HIDMS-PSO-CNN-Attention-RF for optimizing the flow rate of the main nozzle according to an embodiment of the present invention; Figure 2 A structural diagram of a main nozzle according to an embodiment of the present invention; Figure 3 A nozzle core structure diagram of an embodiment of the present invention; Figure 4 A cross-sectional structural diagram of a nozzle core according to an embodiment of the present invention; Figure 5 A three-dimensional flow field network model of a main nozzle according to an embodiment of the present invention; Figure 6 The CNN convolutional neural network architecture of the embodiment of the present invention; Figure 7 It is the convolutional attention module CBAM structure of the present invention; Figure 8 The RF network structure of an embodiment of the present invention; Fig. 9 This is a structural diagram of a CNN-Attention-RF model according to an embodiment of the present invention; Fig.10 It is an INRBO flow chart of an embodiment of the present invention; Fig.11 is the degree of fitting of the training samples in the embodiment of the present invention; Fig.12 This is a schematic diagram of model training in an embodiment of the present invention; Fig.13 A topological structure diagram of a heterogeneous particle swarm according to an embodiment of the present invention; Fig.14 It is a flow chart of the nozzle optimization algorithm of the heterogeneous improved multi-strategy particle swarm algorithm according to an embodiment of the present invention; Fig.15 This is a comparison chart of the results before and after optimization of an embodiment of the present invention.

[0019] Reference numerals 1. Nozzle core; 2. Nozzle body; 3. Conical sleeve; 4. Yarn guide tube; 5. Weft yarn inlet; 6. High-pressure air inlet; 7. Annular air chamber; 8. Conical air chamber. DETAILED DESCRIPTION

[0020] The embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] Example like Figure 2 As shown, the main nozzle is composed of a nozzle core 1, a nozzle body 2, a conical sleeve 3 and a yarn guide tube 4. The high-pressure airflow enters the annular air chamber 7 through the high-pressure air inlet 6 of the main nozzle, enters the conical air chamber 8, the wake area and the subsonic acceleration area from the guide groove, and is ejected from the yarn guide tube 4. The weft yarn enters from the weft yarn inlet 5. The throat of the main nozzle is located in the conical area between the conical small diameter end face of the nozzle core 1 and the inner wall of the yarn guide tube 4. The airflow is accelerated in the throat, and the Mach number reaches or even exceeds 1. In the weft yarn inlet, the dynamic pressure changes from small to large until it reaches the conical negative pressure area, which is conducive to sucking the weft yarn into the yarn guide tube 4. After passing the conical area, the dynamic pressure change area is stable. The airflow outside the yarn guide tube 4 of the main nozzle belongs to a free submerged circular jet. Due to the entrainment and diffusion effects, the velocity of the main nozzle jet is lost along the central axis, resulting in a sharp drop.

[0022] The nozzle core 1 structure can change the jet velocity of the main nozzle. By optimizing the nozzle core 1 structure, the highest jet velocity can be obtained. Figure 3 , Figure 4 The main structural design parameters of the nozzle core 1 include: the outer diameter Dh of the annular air chamber 7, the length Ld of the guide groove, the large end diameter Dz1 of the cone surface of the nozzle core 1, the small end diameter Dz2 of the cone surface of the nozzle core 1, the length Lz1 of the cone section of the nozzle core 1, and the diameter E of the circular hole guide groove. In order to find the optimal structural parameters, the above six design parameters are taken as optimization objects.

[0023] like Figure 1 A main nozzle optimization method based on a hybrid network and a multi-strategy particle swarm algorithm comprises the following steps: S1. Establish the three-dimensional flow field network model of the main nozzle, such as Figure 5The data of the main nozzle design parameters are selected, the internal and external flow field distribution laws corresponding to the three-dimensional flow field network model corresponding to different main nozzle structures are simulated, the central axis velocity curve is extracted, and an experimental device is designed to verify the central axis monitoring point velocity, obtain data samples containing axis velocity and design parameters, and preprocess the data samples.

[0024] The data samples are obtained as follows: 1000 sets of different data are obtained by using advanced Latin hypercube sampling within the corresponding value range of the above six design parameters, and specific parameters such as boundary conditions and initial conditions are set in Fluent or Star ccm+ multi-physics field computational fluid dynamics solver, including temperature, total pressure, static pressure, turbulent kinetic energy, and turbulent kinetic energy dissipation rate. The internal and external flow field characteristics corresponding to the above three-dimensional flow field network model are obtained through CFD numerical simulation, and the corresponding 1000 main nozzle axis velocity curves are obtained.

[0025] The above 1000 sets of design parameters are combined with their corresponding 1000 sets of main nozzle axis speed to obtain 1000 sets of data samples, which are standardized and normalized to the interval [0,1]. The data samples are divided into 78% training set, 15% validation set, and 7% test set. A Pitot tube experimental device is designed to test the main jet axis speed, and the numerical results are verified to construct a validation set for deep learning.

[0026] S2. Build a CNN-Attention-RF neural network model.

[0027] In the CNN-Attention-RF neural network model, the convolutional neural network CNN is first used to perform convolution processing on the input parameter data of the main nozzle shape structure and the output axis jet velocity to better extract new data features and enhance the correlation between input data and output data. Then the attention mechanism Attention is used to weightedly fuse the features extracted by CNN so that the network can automatically focus on information that is more useful for prediction. The random forest regression model RF is introduced to balance the errors of unevenly distributed data in the sample, and has a good tolerance for outliers and noise. RF can not only process large multidimensional data, but also is suitable for nonlinear problems.

[0028] The prediction method based on the CNN-Attention-RF hybrid neural network combines three network structures: convolutional neural network (CNN), attention mechanism (Attention) and random forest (RF) to improve the accuracy and efficiency of model prediction.

[0029] The model architecture of the CNN-Attention-RF hybrid neural network is as follows: Fig. 9 shown.

[0030] like Figure 6 Input layer: includes the number of data samples contained in a batch, the number of time steps contained in each data sample, and the feature dimension of the input data.

[0031] Convolutional layer (CNN): Use 2D convolutional layer to transform the input feature dimension into a higher feature dimension. After the convolution operation, the output shape is

[0032] The calculation of the convolution layer can be expressed as: Z = Conv2D (X), where the shape of Z is (batch_size, 64, sequence_length), which means that for each input sample, the convolution layer generates 64 feature maps, and the length of each feature map is the same as the length of the input sequence. These feature maps capture different feature patterns in the input sequence for further processing by subsequent layers.

[0033] CNN is mainly used to capture local features in sequence data. The multi-dimensional features of the simulation data are input and the features are extracted through the convolution layer.

[0034] Attention layer: Apply the Attention mechanism to weight the output of the CNN so that the model focuses on different parts of the input sequence.

[0035] like Figure 7 As shown. CBAM mechanism: The specific calculation of the attention mechanism of the convolution module can be abstracted into three stages: In the first stage, the global max pooling and global average pooling of width and height are respectively passed, and the output features of MLP are respectively passed, and the channel attention features and input features are element-wise weighted, and then passed through the sigmoid activation function to generate the channel attention features. The channel attention features and input features are element-wise multiplied to generate the input features of the spatial attention module.

[0036] The second stage performs spatial attention mechanism operations, compresses the channels, and performs average pooling and maximum pooling in the channel dimension.

[0037] In the third stage, the previously extracted features are merged into a 2-channel feature to generate the output feature.

[0038] The relevant formula is as follows: ; ; in, is the channel attention weight, is the spatial attention weight, The convolutional layer is average pooled. Perform max pooling on the convolutional layer.

[0039] The attention mechanism is used to perform weighted fusion of the features extracted by CNN so that the network can automatically focus on information that is more useful for prediction.

[0040] like Figure 8 As shown. In the third process, the feature vector generated by the CNN feature extraction network is used as the input feature of the RF regressor according to the random forest, and is divided into a test set and a training set. The bootstrap sampling method in RF is used to sample the training set, generate the corresponding training subset for each decision tree, and train the decision tree. The optimal attribute is selected at each internal node for splitting to generate the final output.

[0041] S3. Optimize the hyperparameters of the CNN-Attention-RF neural network model based on the improved INRBO algorithm.

[0042] An improved algorithm based on the Newton-Raphson optimizer (NRBO) is proposed as INRBO.

[0043] The NRBO optimization algorithm is based on two key mechanisms: the Newton-Raphson search rule (NRSR) and the trap avoidance operator (TAO). NRSR improves the algorithm's search capability and convergence speed by using first-order and second-order derivatives to speed up the update of positions. TAO helps the algorithm avoid falling into local optimal solutions by increasing the randomness and diversity of solutions.

[0044] like Fig.10 As shown. The INRBO algorithm specifically includes the following steps: 1) Initialization parameters, including population size, maximum number of iterations, difference factor, upper and lower limits of hyperparameter vectors.

[0045] 2) Initialize the population using a chaotic sequence.

[0046] Specifically: The use of Tent chaotic mapping to initialize the particle positions can enhance the diversity of the initial population and prevent the algorithm from falling into the local optimum in the initial stage. By introducing chaotic sequences, more random and unpredictable initial positions are generated, improving the algorithm's exploration ability.

[0047] For each particle position , randomly generated in the interval ([0, 1]): ∼Uniform(0,1); Apply the Tent map to each randomly generated value: ; The chaotic sequence x' Map to the specified search space range: ; in, UB is the upper bound of the search space, LB is the lower bound of the search space, d It is the parameter of Tent mapping.

[0048] 3) Adaptive inertia weight calculation, using dynamically adjusted inertia weight. This weight will gradually change with the number of iterations. In the initial stage of the algorithm, the inertia weight is large, tending to global search, thereby increasing the exploration space; in the later stages of the iteration, the inertia weight decreases, tending to local development. This strategy helps to explore the solution space extensively in the early stage and converge to the optimal solution in the later stage.

[0049] Inertia weight formula: ; in, is the inertia weight of the current iteration, is the current iteration number, is the maximum number of iterations, is the maximum value of the inertia weight, is the minimum value of the inertia weight, α To control the steepness of the Sigmoid function, β To control the center position of the Sigmoid function.

[0050] 4) Apply NRSR search and adjust the particle position update according to the difference between the optimal solution and the worst solution.

[0051] NRSR searches specifically include: calculate ΔX , ; Calculate the first NRSR, , rand is a random number sampled from a standard normal distribution.

[0052] Introducing an exponential decay factor , , linear attenuation factor

[0053] ; ; ; use , , Three factors dynamically generate parameters

[0054] ; Randomly select four individual indices for difference, combined with parameters , update the location , ; calculate and , ; Calculate the second NRSR, ; in, ΔX is a random perturbation used to adjust the current solution, NRSR is a search rule based on random perturbations. Location Updates To select different update strategies based on the value of the position, , They are respectively combined by random weights Position, Generate new search directions. The second NRSR is to further adjust the search direction and increase the diversity and flexibility of the search.

[0055] 5) Determine whether the random number is less than the difference factor. If the random number is less than the difference factor, perform TAO update. If the random number is greater than or equal to the difference factor, perform local optimal detection. When the local optimal detection detects that the improvement of the optimal solution is less than the preset threshold, the stagnation counter is increased. When the stagnation counter reaches a certain number of times, it is determined that the current search is trapped in the local optimal state, thereby triggering chaotic perturbation. Use the chaotic perturbation Logistic mapping to update the position of the solution, apply perturbation to the current position, so that the particle jumps out of the local optimal area and restores the global search capability.

[0056] Iterate the Logistic map for 10 times to enhance the chaotic effect: .

[0057] Logistic Mapping Parameters r Take 4.

[0058] The number of local optimal detections is limited to 10, and the improvement threshold is 1×10 -6 .

[0059] 6) Determine whether the iteration is terminated. If the iteration is terminated, update the convergence curve to record the best result and output the result.

[0060] S4. INRBO-CNN-Attention-RF model predicts the axial airflow velocity of main nozzles with different structures through training.

[0061] The INRBO algorithm is used to optimize the hyperparameters of the CNN-Attention-RF model: convolution kernel size, number of neurons in each convolution layer, number of samples per iteration and training rounds, number of trees in the random forest, maximum depth of the tree, minimum number of samples required for each internal node, minimum number of samples required for each leaf node, and learning rate. First, the parameters of the INRBO algorithm, maximum number of iterations, population size, deep learning epoch, objective function, minimum and maximum ranges of hyperparameter vectors are set. The number of batches per CNN1 [32, 128], convolution kernel size [1, 2]; the number of batches per CNN2 [16, 64], convolution kernel size [1, 2]; the number of random forest trees [10, 200], maximum depth of the tree [3, 30], minimum number of samples per internal node [2, 10], minimum number of samples required for each leaf node [1, 4] are configured to build and train the hybrid network model, and the root mean square error (RMSE) is used as the loss function for the regression task. RMSE measures the difference between the predicted value and the actual value to find the optimal hyperparameters of the hybrid model.

[0062] ; The performance of the hybrid network model is measured by mean absolute error (MAE), root mean square error (RMSE), coefficient of determination (R 2 ) are evaluated to determine the effectiveness of each parameter configuration.

[0063] Through multiple iterations (multiple epochs, forward propagation in each epoch, calculating output, calculating loss, backpropagation, and updating model parameters), the INRBO algorithm gradually optimizes the hybrid model hyperparameters. In each iteration, the algorithm evaluates the performance of the current parameter settings and monitors the loss function (RMSE) and evaluation indicators (R 2 , MAE) on the validation set to evaluate the performance of the hybrid network model. And adjust it according to the natural rhythm strategy to explore a better solution. Finally, the hybrid CNN-Attention-RF network is constructed using the optimized hyperparameter configuration. The parameter settings of the CNN-Attention-RF hybrid network model optimized by INRBO are shown in Table 1.

[0064] Table 1 Parameter settings of INRBO-CNN-Attention-RF hybrid network model ;

[0065] The training sample fitting degree of the CNN-Attention-RF hybrid network model optimized by INRBO is as follows: Fig.11 As shown in Fig.12 shown.

[0066] Using simulation data, the hyperparameters of the CNN-Attention-RF model are optimized by the improved INRBO algorithm. The comparison of the evaluation indicators of the INRBO-CNN-Attention-RF, CNN, RF and CNN-Attention prediction models is shown in Table 2.

[0067] Table 2 Comparison of evaluation indicators of different prediction models ;

[0068] The results show that compared with the single CNN network or RF machine learning regression model and the CNN-Attiention deep learning model, the CNN-Attention-RF hybrid model has smaller prediction error and higher prediction accuracy for the jet velocity on the main nozzle axis.

[0069] The trained hybrid network can be used for speed prediction. When new structural parameter data is input, the network will output the corresponding speed prediction value.

[0070] S5. Obtain the optimal structural parameters of the main nozzle based on the multi-strategy particle swarm algorithm.

[0071] The agent model INRBO-CNN-Attention-RF is constructed by combining deep learning with machine learning. The airflow velocity of the main nozzles with different structures is predicted through training. Based on the trained prediction set, the multi-strategy particle swarm algorithm HIDMS-PSO based on heterogeneous improvement is used to find the best output and the highest outlet velocity according to the given input value range. The main nozzle structure parameters corresponding to the input are the optimal structure. The HIDMS-PSO algorithm flow chart is as follows Fig.14 shown.

[0072] like Fig.13 As shown. Heterogeneous improved multi-strategy particle swarm algorithm HIDMS-PSO is an advanced meta-heuristic optimization algorithm based on particle swarm optimization algorithm (PSO). The algorithm utilizes heterogeneous inertia weights and dynamic multi-subgroup strategies to improve search efficiency and convergence performance. Specifically, HIDMS-PSO divides the particle swarm into several units, each of which contains a master particle and three slave particles. Particles in different units not only interact with each other, but also exchange information with other units to avoid premature convergence.

[0073] The assumed distinction between subordinate particles and the unit structure hierarchy between particles is used to restrict and control the information flow between particles, rather than a global or arbitrary information exchange. The communication model can be summarized by the following rules: 1. The principal particle of the unit is not directly related to the j1. The master particles of a unit can communicate only through slave particles. 2. The master particles can only exchange information with slave particles within the unit. 3. Slave particles can only communicate with other unit slave particles of the same type, so they cannot communicate with slave particles within the unit.

[0074] (1) Classical Particle Swarm Optimization The group consists of N particles, and the speed of each particle is , at time t The location is , the best position of an individual is pbest , the global best position gbest The particles are n By using three vectors in the , pbest and gbest , that is, the position of the global best particle in the group. The standard PSO algorithm uses the following two equations to update the velocity and position of the particle: ; ; in ω is the inertia weight, affecting the inertia of the particle. c 1 and c 2 is the acceleration coefficient, which controls the particles to fly towards pbest and gbest speed. R 1 and R 2 is a random vector ∈ [0,1].

[0075] (2) Search method In the HIDMS-PSO algorithm, the initial population is divided into two equal subpopulations, one is homogeneous and the other is heterogeneous, and each subpopulation adopts a different movement strategy. The homogeneous subpopulation uses the update equation of the classic PSO algorithm, while the heterogeneous subpopulation forms N The inward-facing strategy aims to guide particles to the same unit. In contrast, the outward-facing strategy uses samples from different units to guide particles.

[0076] Inward Moving Strategy: The inward moving strategy uses the position information obtained from all unit members to provide guidance for particles. This strategy essentially allows each unit to explore and exploit local solutions simultaneously.

[0077] The master particles randomly choose any of the following three equations to update their velocity: ; ; ; in: is the main particle velocity, The main particle at time t location, For the n The particle with the smallest position similarity among the units, It is n The unit is best suited for the particle position, is the average position of all slave particles in the current unit of the master particle; The slave particles are moved using the following formula: ; in, is the particle velocity, p best For a particle at time t The best location found, is the particle position, For the n The main particle position of each unit; Outward-Moving Strategy: Compared with the inward-moving strategy, the outward-moving strategy provides particles with guidance from different units in the population while maintaining the hierarchical structure.

[0078] The master particles randomly choose any of the following three equations to update their velocity: ; ; ; in, is the main particle velocity, For the n The average position of particles in a cell, is the position of the primary particle of the randomly selected unit, is the average position of the particle's own unit members; A slave particle moves toward a random slave particle of the same type belonging to another cell using the following velocity update equation: .

[0079] By combining these two strategies and allowing a part of the population to perform classic PSO position updates, the entire population is divided into a homogeneous group and a heterogeneous group. The heterogeneous population is further divided into N pUnits are needed for efficient exploration and exploitation. Homogeneous populations focus on a single solution output, while heterogeneous populations focus on multiple solutions.

[0080] The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm described in the present invention is used to obtain the optimal parameter design. The comparison between the optimal design parameters and other parameters is shown in Table 3.

[0081] Table 3 Comparison of optimal structural parameter design with other parameters ;

[0082] As can be seen from Table 3, under the same gas supply pressure of 0.4 MPa, the maximum velocity along the main jet axis was obtained through numerical simulation and Pitot tube experimental test for main nozzles with different structures. As can be seen from Table 3, group 1 is the optimized main nozzle structure parameters, and the maximum outlet velocity is 348.42m / s, which is higher than the jet velocity of other group main nozzles, and the flow rate is 1.71m 3 / h, which is less than the main nozzles of group numbers 2, 3, 4, and 5. The results in Table 3 show that under the same pressure conditions, the main nozzle structure optimized by the HIDMS-PSO algorithm significantly increases the jet velocity along the axis, which reduces the gas consumption to a certain extent. The production unit can use the optimized main nozzle on the air jet loom to improve the weft insertion efficiency and reduce the air consumption per unit time, thereby reducing the consumption of electricity and the production cost of weaving.

[0083] Fig.15 1 is a comparison chart of the results before and after optimization of an embodiment of the present invention. Fig.13 It can be seen that after the main nozzle parameters are optimized, the outlet jet velocity of the main nozzle is significantly improved and the outlet flow rate is reduced.

[0084] Therefore, the main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm described in the present invention is adopted, and a regression model of main nozzle mechanism parameters and speed is constructed by combining deep learning neural network with machine learning algorithm. The improved INRBO optimization algorithm is used to optimize the hyperparameters of the CNN-Attention-RF regression model. The improved model is used to predict the speed of the main nozzle jet at different points along the axis. Then the HIDMS-PSO algorithm is used to find the main nozzle structural parameters corresponding to the highest speed predicted by the proxy model. This can improve the weft insertion efficiency of the air-jet loom and reduce the R&D cost and cycle, which has important theoretical significance and engineering value.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm, characterized in that: The following steps are involved: S1. Establish a three-dimensional flow field numerical model of the main nozzle, select the data of the main nozzle design parameters, simulate the internal and external flow field distribution laws of the three-dimensional flow field model corresponding to different main nozzle structures, extract the central axis velocity curve, and design an experimental device to verify the central axis monitoring point velocity, obtain data samples containing the axis velocity and the main nozzle structure design parameters, and pre-process the data samples; S2. Build a CNN-Attention-RF hybrid model; S3. Optimize the hyperparameters of the CNN-Attention-RF hybrid network model based on the INRBO algorithm; S4, INRBO-CNN-Attention-RF model predicts the airflow velocity along the axis of the main nozzle with different structures through training; S5. Based on the multi-strategy particle swarm optimization algorithm, the optimal structural parameters of the main nozzle are obtained to increase the jet speed of the main nozzle, thereby improving the weft insertion efficiency.

2. The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm according to claim 1 is characterized in that: In S1, the data samples are obtained specifically as follows: according to the main nozzle design parameters, a number of different structure data are sampled using advanced Latin hypercube sampling within the corresponding value range, boundary conditions and initial conditions are set using a general solver of computational fluid dynamics, and the velocity along the axis of the main nozzle jet is obtained through numerical simulation using the general solver of computational fluid dynamics, training set and test set samples for deep learning are constructed, and a Pitot tube experimental device is designed to test the axis velocity of the main jet, the numerical results are verified, and a verification set for deep learning is constructed.

3. The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm according to claim 2 is characterized in that: In S2, in the CNN-Attention-RF neural network model, the convolutional neural network is used to perform convolution processing on the main nozzle input structural parameter data and the output jet axis velocity, extract new data features, and enhance the correlation between the input data and the output data; The attention mechanism is used to perform weighted fusion on the features extracted by the convolutional neural network so that the network can automatically focus on information that is more useful for prediction; the random forest regression model balances the errors of unevenly distributed data in the samples and performs regression prediction.

4. The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm according to claim 3 is characterized in that: In S3, the INRBO algorithm specifically includes the following steps: 1) Initialization parameters; 2) Initialize the population using chaotic sequences; 3) Adaptive inertia weight calculation, using dynamically adjusted inertia weight; 4) Apply the Newton-Raphson search strategy to adjust the position update of the vector according to the difference between the optimal solution and the worst solution; 5) Determine whether the random number is less than the difference factor; if the random number is less than the difference factor, perform TAO update; if the random number is greater than or equal to the difference factor, perform local optimal detection. When the local optimal detection detects that the improvement of the optimal solution is less than the preset threshold, increase the stagnation counter. When the stagnation counter reaches a certain number of times, it is determined that the current search is trapped in the local optimal state, triggering chaotic perturbation. Use chaotic perturbation Logistic mapping to update the position of the solution, apply perturbation to the current position, so that the particle jumps out of the local optimal area and restores the global search capability. 6) Determine whether the iteration is terminated. If the iteration is terminated, update the convergence curve to record the best result and output the result.

5. The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm according to claim 4 is characterized in that: In 2), the population is initialized using a chaotic sequence as follows: The position of the initialization vector is initialized using the Tent chaotic map. By introducing a chaotic sequence, a more random and unpredictable initialization position is generated, which improves the exploration ability of the algorithm. For each particle position , randomly generated in the interval ([0,1]): ∼Uniform(0,1); Apply the Tent map to each randomly generated value: ; The chaotic sequence x' Map to the specified search space range: ; in, UB is the upper bound of the search space, LB is the lower bound of the search space, d It is the parameter of Tent mapping.

6. The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm according to claim 5 is characterized in that: In the above 3), Inertia weight formula: ; in, is the inertia weight of the current iteration, is the current iteration number, is the maximum number of iterations, is the maximum value of the inertia weight, is the minimum value of the inertia weight, To control the steepness of the Sigmoid function, β To control the center position of the Sigmoid function.

7. The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm according to claim 6 is characterized in that: In 4), NRSR is a search rule based on random perturbations, specifically including: calculate ΔX , ; Introducing three operators, modifying Parameters guide the population to update in the correct direction. The expression is: ； ； ； ； is a random number that follows a standard normal distribution. r 1 , r 2 、r 3 , r 4 are four individual indices randomly selected from the population for differential calculation. X b is the best vector position, is the current position of the vector, F 1 , F 2 is the exponential decay factor, F 3 is the linear attenuation factor; Calculate the first NRSR, ; Update Location , ; calculate and , ; Calculate the NRSR of the second one, ； in, ΔX is a random perturbation used to adjust the current solution, r 1 , r 2 are different integers randomly selected from the population; NRSR is a search rule based on random perturbations. Location Updates To select different update strategies based on the value of the position, y w , y b They are respectively combined by random weights Position, Generate new search directions, The second NRSR is to further adjust the search direction and increase the diversity and flexibility of the search.

8. The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm according to claim 7 is characterized in that: In S5, the multi-strategy particle swarm algorithm divides the particle swarm into several units, each unit includes a master particle and several slave particles, and the particles in the same unit interact with each other and exchange information with other units.

9. The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm according to claim 8, characterized in that: The initial population of the multi-strategy particle swarm algorithm is divided into two equal isomorphic subgroups and heterogeneous subgroups. The isomorphic subgroup uses the classic PSO algorithm to update the equation, and the heterogeneous subgroup forms N p The inward moving strategy guides particles to the same unit, and the outward moving strategy uses samples from different units to guide particles.

10. The main nozzle optimization method based on hybrid network and multi-strategy particle swarm algorithm according to claim 9, characterized in that: The inward movement strategy uses the position information obtained from all unit members to provide guidance for the particles. The master particles randomly choose any of the following three equations to update their velocity: ； ； ； in: is the main particle velocity, The main particle at time t location, For the n The particle with the smallest position similarity among the units, It is n The unit is best suited for the particle position, is the average position of all slave particles in the current unit of the master particle, ω is the inertia weight, constant c 1 and c 2 is the acceleration factor, R 1 and R 2 is a random vector ∈ [0,1]; The slave particles are moved using the following formula: ； in, is the particle velocity, p best For a particle at time t The best location found, is the particle position, For the n The main particle position of each unit; The outward movement strategy provides particles with guidance from different units in the population while maintaining the hierarchy. The master particles randomly choose any of the following three equations to update their velocity: ； ； ； in, is the main particle velocity, For the n The average position of particles in a cell, is the position of the primary particle of the randomly selected unit, is the average position of the particle's own unit members; R 1 and R 2 is a random vector ∈ [0,1]; A slave particle moves toward a random slave particle of the same type belonging to another cell using the following velocity update equation: .

Images