Aviation fuel centrifugal pump multi-objective optimization method based on artificial bee colony algorithm

Abstract

The invention discloses an aircraft fuel centrifugal pump multi-objective optimization method based on an artificial bee colony algorithm. The method comprises the steps of 1, building a BP neural network; 2, programming an artificial bee colony algorithm, and optimizing centrifugal pump output parameters by taking the BP neural network in the step 1 as a black box function; 3, programming through the artificial bee colony algorithm to obtain a final nectar source; 4, performing three-dimensional modeling on the input parameter (namely a nectar source) optimized by the artificial bee colony algorithm, performing CAE simulation analysis through related software, comparing a simulation prediction result with a neural network output result, and comparing a lift error and an efficiency error; and the optimized result is compared with the result before optimization, and the lift and the efficiency improvement value are compared. The method has the characteristics of simplicity, high efficiency and short time consumption.

Description

Technical Field

[0001] The invention relates to the technical field of multi-objective optimization of aviation fuel centrifugal pumps, and in particular to a multi-objective optimization method of aviation fuel centrifugal pumps based on an artificial bee colony algorithm. Background Art

[0002] Aircraft engine fuel centrifugal pumps are one of the common power sources for aviation equipment. Their main function is to inject fuel at a certain flow rate and pressure into the combustion chamber of the aircraft engine to ensure that the engine can generate sufficient thrust. They are composed of an impeller and a volute, and usually adopt a centrifugal working principle, driving the working fluid to rotate through the rapid rotation of the impeller blades. The working fluid is accelerated radially and decelerated in the volute part, thereby converting kinetic energy into pressure energy. The impeller provides two forms of energy to the liquid: static pressure energy and kinetic energy.

[0003] Traditional centrifugal pump design is usually based on experience and trial and error methods, such as using empirical formulas to reflect the mapping relationship between centrifugal pump performance and various influencing factors to obtain the optimal value of centrifugal pump efficiency. After the popularization of numerical analysis software, the design of fuel centrifugal pumps relies on computational fluid dynamics (CFD) simulation and test data, and then uses numerical optimization calculation methods under constraints. It has the advantages of saving time and cost, predictability, etc., and has become an indispensable software in the fluid simulation process.

[0004] However, traditional design methods rely on experience and trial and error, which may lead to suboptimal design solutions. In addition, these methods usually fail to consider the trade-offs between multiple performance objectives. Traditional CFD simulation methods usually require a lot of computing resources and time, especially in multi-objective optimization design, the computational cost is higher, and they may be affected by numerical instability, which may lead to inaccurate results or the collapse of numerical solutions.

[0005] The difficulty of multi-objective optimization of centrifugal pumps lies in how to determine the correct objective function during optimization. Due to the complexity of the flow field in the pump, it is currently impossible to strictly derive the relationship between the design parameters and the optimization objectives through mathematical formulas. If CFD technology is used to solve the value of the objective function required by the optimization algorithm, this process may require hundreds or even tens of thousands of simulation tasks, which will take a lot of time.

[0006] When the experimental design method is used for optimization, the analysis within the selected level is more accurate, but the area outside the level range cannot be accurately analyzed.

[0007] In summary, the defects of the prior art are that the relationship between the design parameters and the optimization objectives cannot be derived by mathematical models and the optimization takes a long time. Summary of the invention

[0008] In order to overcome the above technical problems, the purpose of the present invention is to provide a multi-objective optimization method for an aviation fuel centrifugal pump based on an artificial bee colony algorithm, which is simple, efficient and time-saving.

[0009] The technical solution adopted by the present invention is:

[0010] A multi-objective optimization method for an aviation fuel centrifugal pump based on an artificial bee colony algorithm comprises the following steps;

[0011] Step 1: Build a BP neural network;

[0012] Step 2: Programming the artificial bee colony algorithm and using the BP neural network described in step 1 as a black box function to optimize the output parameters of the centrifugal pump;

[0013] Step 3: Obtain the final nectar source through the artificial bee colony algorithm programming;

[0014] Step 4: Carry out 3D modeling of the input parameters (i.e. honey source) optimized by the artificial bee colony algorithm and perform CAE simulation analysis through relevant software, and compare the simulation prediction results with the output results of the neural network. After the optimization of the artificial bee colony algorithm, the centrifugal impeller head increased by 27.67m and the efficiency increased by 0.82%.

[0015] The step 1 is specifically as follows:

[0016] A BP neural network is established based on existing centrifugal pump samples, and the prediction accuracy is improved through experimental data training and verification to adapt to fuel centrifugal pumps of different types and sizes, and accurately predict the performance parameters of fuel centrifugal pumps at multiple operating points, including but not limited to head, efficiency, etc.

[0017] For the parameter optimization of aviation centrifugal pumps, the construction of BP neural network specifically includes the following processes:

[0018] (1) Data preparation: preparing data sets for training and testing neural networks; the present invention collects data of 50 sets of aviation centrifugal pumps;

[0019] The input parameters, namely the BP neural network input layer, include the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle

[0020] The output parameters, i.e. the output layer of the BP neural network, include the centrifugal pump head H and efficiency η. The 50 sets of data are divided into 35 training sets, 8 prediction sets and 7 validation sets. The corresponding input layer and output layer matrix sizes are (50*5) and (2*50) respectively.

[0021] (2) Determine the network architecture: The constructed BP neural network includes an input layer, a hidden layer and an output layer. The input layer has 5 parameters. After comparing the network prediction results of different numbers of nodes in the hidden layer, it is finally determined that when the hidden layer nodes are 10, the fitting black box function is optimal, so 10 hidden layer nodes are selected. The output layer has 2 output parameters, corresponding to the centrifugal pump head and efficiency.

[0022] (3) Initialize weights and thresholds: Use the rand function in the programming software to randomly initialize W1 and b1;

[0023] Where: W1 is the weight, b1 is the bias;

[0024] (4) Selecting the activation function: The activation function of the neural network uses the Sigmoid function. The Sigmoid function is a nonlinear function that allows the neural network to learn nonlinear relationships. Multiple neurons in the neural network can represent complex nonlinear functions through the combination of Sigmoid functions. They have smooth gradients in most of their input ranges, which is important for the stability and convergence of the gradient descent optimization algorithm. Therefore, the sigmoid function is selected as the activation function of the neural network, and the output range is mapped to between 0 and 1. The function form is:

[0025]

[0026] (4) Forward propagation: This design passes the input data into the network and calculates the output of each neuron through the neurons in each layer;

[0027] The input layer is represented by L1, the output layer is represented by L2, the weight between L1 and the hidden layer is represented by V, and the matrix size is 50*5; the weight between L2 and the hidden layer is represented by W, and the matrix size is 2*50;

[0028] (5) Backpropagation: Update the parameters by calculating the gradient of the loss function with respect to the weights and biases to reduce the value of the loss function. Backpropagation uses the gradient descent algorithm or its variants to adjust the parameters to minimize the loss.

[0029] The step 2 is specifically as follows: based on the artificial bee colony algorithm, the three roles of leading bees, following bees and scout bees are used to generate new design parameter combinations, and the performance of each nectar source is evaluated, and the optimal solution is determined by comparing the amount of nectar in each place. After multiple iterations of the artificial bee colony algorithm, the optimal solution for food is continuously updated.

[0030] This cyclic optimization process is used to search for the best design parameters of nearby honey sources, including but not limited to the impeller geometry, the inlet and outlet placement angles of the front and rear cover blades, etc., so as to maximize or minimize multiple performance and constraint objectives.

[0031] The specific operation of designing parameters in step 2 is:

[0032] S1 defines the fitness function fit: In the multi-objective optimization process, the centrifugal pump head is required to meet the given head requirement while improving the centrifugal pump efficiency and optimizing the input structural parameters, that is, the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle And to make the centrifugal pump most efficient, the output results are screened through greedy comparison of conditional statements in the fitness function, and the mathematical model of the theoretical head of the centrifugal pump is imported for calculation. When the head reaches the requirement, the efficiency of the centrifugal pump is output and the input parameters corresponding to the maximum efficiency are screened as the optimal solution;

[0033] S2 Artificial Bee Colony Algorithm Parameter Setting: Establish a zero matrix to store the initial honey source array; the so-called initial honey source is to store the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle Column vector of the five input parameters.

[0034] In the present invention, the number of nectar sources is set to 100, the number of leading bees is set to 100, the number of iterations is set to 200, the acceleration coefficient is set to 1, and the random number is constrained between the set minimum and maximum structural parameter values ​​using the following formula in the randomly found nectar source and the optimal solution is searched in this interval:

[0035] zzza=unifrnd(varmin1,varmax1);

[0036] zzzb=unifrnd(varmin2,varmax2);

[0037] zzzc=unifrnd(varmin3,varmax3);

[0038] zzzd=unifrnd(varmin4,varmax4);

[0039] zzze=unifrnd(varmin5,varmax5);

[0040] pop(i).position=[zzza zzzb zzzc zzzd zzze];

[0041] pop(i).cost=sim(net,poo(i).position′);

[0042] The first five rows represent the continuous and uniform generation of random numbers within the range of optimized parameter values;

[0043] The 6th row represents placing the randomly generated parameters in a space;

[0044] Line 7 represents substituting the generated parameters into the black box function to calculate the result;

[0045] S3 Artificial Bee Colony Algorithm Programming: Artificial Bee Colony Algorithm iteration includes initialization, discovery, information exchange, and abandonment;

[0046] The initialization process is to randomly generate a certain number of initial bee individuals, each of which represents a potential solution;

[0047] The discovery process is that the leading bee randomly searches for a new nectar source near nectar source i. When the fitness of the newly found high-quality nectar source is greater than that of the old nectar source, the high-quality nectar source will be selected to replace the original nectar source.

[0048] The information exchange process adopts the roulette method, which puts the structural parameter combination of each nectar source into the fitness function for calculation, and returns the calculated lift and efficiency. The probability of a follower bee selecting a leader bee in the roulette is the ratio of the efficiency return value of the nectar source to the sum of the efficiency return values ​​of all nectar sources.

[0049] The abandonment process is that when the trial iterative search approaches the threshold L in the search process, if the nectar source has not been updated to a better nectar source, the nectar source will be abandoned, and the scout bee will jump out of the nearby nectar source range and find a new nectar source;

[0050] S4 iteration: Repeat the iteration process. Since the fitness function contains the bp neural network to predict the efficiency of each nectar source, when running the artificial bee colony algorithm, it is necessary to first run the bp neural network to establish a prediction function for the input and output parameters. After iteration, the optimal solution is obtained.

[0051] The initialization process in S3 is to randomly generate a certain number of initial bee individuals, each bee represents a potential solution;

[0052] The discovery process in S3 is that the leading bee randomly searches for a new nectar source near nectar source i. When the fitness of the newly found high-quality nectar source is greater than that of the old nectar source, the high-quality nectar source will be selected to replace the original nectar source.

[0053] The information exchange process in S3 adopts the roulette method, puts the structural parameter combination of each nectar source into the fitness function for calculation, and returns the calculated lift and efficiency. The probability of a follower bee selecting a leading bee in the roulette is the ratio of the efficiency return value of the nectar source to the sum of the efficiency return values ​​of all nectar sources;

[0054] The abandonment process in S3 is that when the trial iterative search approaches the threshold L in the search process, if the nectar source has not been updated to a better nectar source, the nectar source will be abandoned, and the scout bee will jump out of the nearby nectar source range and find a new nectar source.

[0055] The step three is specifically as follows: after multiple iterations of nectar source updates, the best fitness function in each iteration process is selected as the output, and when the output fitness function converges, it is the optimal solution.

[0056] The corresponding example of the present invention needs to optimize the efficiency according to the initial centrifugal pump parameters, where the required head needs to meet the head of 1685m, β s1 =48.9°, β h1 =60.8°β s2 =12.4°, β h2 =14.0°, The structural parameters to be optimized have an efficiency of 75.90%. After optimization, the lift meets the requirements and the efficiency is maximized;

[0057] After consulting the data, the optimization range of the input parameters of the artificial bee colony algorithm is: β s1 =46~51,β h1 =57~62β s2 =11~14,β h2 =14.5~17.5, Optimize the input parameters within this range. The following is the specific operation process:

[0058] First run the BP neural network in the programming software, and its training model regression is as follows:

[0059] The R values ​​of the training set, validation set, and test set are 0.99961, 0.99855, and 0.99956, respectively, and the total R value is 0.99942. It can be considered that this black box function has high reliability.

[0060] Run the artificial bee colony algorithm to perform iterative processes such as initialization, discovery, information exchange, and abandonment;

[0061] After multiple iterations, the fitness function tends to be stable, and the efficiency at this time can be considered to be the optimal value and the optimized centrifugal pump parameters are: β s1 =57.2°, β h1 =46°, β s2 =15.7°, β h2 =14.0°,

[0062] The step 4 specifically includes: performing three-dimensional modeling on the input parameters optimized by the artificial bee colony algorithm and performing CAE simulation analysis through relevant software, and comparing the simulation prediction results with the output results of the neural network;

[0063] S1 3D modeling: 3D modeling is performed using relevant software, mainly modeling the fluid domain of the impeller and volute;

[0064] S2 simulation analysis: The simulation analysis is carried out through relevant software, mainly simulating the inlet and outlet pressures and calculating the lift. The formula is as follows;

[0065]

[0066] Where P out is the centrifugal pump outlet pressure, P in is the inlet pressure of the centrifugal pump, and γ is the bulk density of the conveying medium.

[0067] Simulate the shaft power of the 3D model and calculate the efficiency of the centrifugal pump. The formula is as follows:

[0068]

[0069] Where P is the shaft power of the centrifugal pump and Q is the design flow rate of the centrifugal pump.

[0070] Beneficial effects of the present invention:

[0071] The present invention provides a multi-objective optimization design method for an aviation fuel centrifugal pump based on an artificial bee colony algorithm. Since the present invention adopts a BP neural network to express a black box function between the two and predict the output parameters, the artificial bee colony algorithm can be used to optimize the output parameters according to the output results within a certain range. The intelligent algorithm and optimization technology are used to reduce the time cost, improve the optimization efficiency and realize multi-objective optimization design for the design optimization of the aviation fuel centrifugal pump.

[0072] The present invention firstly builds a BP neural network, and then programs an artificial bee colony algorithm and uses the BP neural network built in the first step as a black box function to optimize the output parameters of the centrifugal pump, and then obtains the final nectar source through the artificial bee colony algorithm programming, and finally three-dimensionally models the input parameters (i.e., the nectar source) after optimization by the artificial bee colony algorithm and performs CAE simulation analysis through relevant software, and compares the simulation prediction results with the output results of the neural network. The head error and efficiency error of the simulation prediction results and the output results of the neural network are 0.90% and 0.55% respectively, and it can be considered that the constructed model is accurate enough; and then the optimized results are compared with the results before optimization, and the head and efficiency are increased by 27.67m and 0.82% respectively, and the improvement percentages are 1.64% and 1.08% respectively, thereby realizing multi-objective optimization of the head and efficiency of the centrifugal pump. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 This is a schematic diagram of the overall process.

[0074] Figure 2 Schematic diagram of the BP neural network propagation process.

[0075] Figure 3 This is the regression diagram of the BP neural network training model.

[0076] Figure 4 This is a diagram of the iterative process of the artificial bee colony algorithm.

[0077] Figure 5 This is a graph showing the results of running the artificial bee colony algorithm.

[0078] Figure 6 This is the fluid domain structure diagram of the aviation fuel centrifugal pump.

[0079] Figure 7 This is the pressure cloud diagram of the optimized aviation fuel centrifugal pump. DETAILED DESCRIPTION

[0080] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0081] like Figure 1-7 As shown: A multi-objective optimization design method for aviation fuel centrifugal pump based on artificial bee colony algorithm, comprising the following steps;

[0082] Step 1: Build a BP neural network

[0083]

[0084]

[0085] Based on the existing centrifugal pump samples (as shown in the table above), a BP neural network is established, and the prediction accuracy is improved through a large amount of experimental data training and verification to adapt to different types and sizes of fuel centrifugal pumps. The characteristic is that the performance parameters of the fuel centrifugal pump can be accurately predicted at multiple operating points, including but not limited to head, efficiency, etc. For the parameter optimization of aviation centrifugal pumps, the construction of the BP neural network specifically includes the following processes:

[0086] Data preparation: Prepare data sets for training and testing neural networks. The present invention collects 50 sets of data on aviation centrifugal pumps, where the input parameters, i.e., the BP neural network input layer, include the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle The output parameters, i.e. the output layer of the BP neural network, include the centrifugal pump head H and efficiency η. The 50 sets of data are divided into 35 training sets, 8 prediction sets and 7 validation sets. The corresponding input layer and output layer matrix sizes are (50*5) and (2*50) respectively.

[0087] Determine the network architecture: The BP neural network constructed in the present invention includes an input layer, a hidden layer and an output layer, and the input layer has 5 parameters; the hidden layer compares the network prediction results of different numbers of nodes, and finally determines that when the hidden layer nodes are 10, the fitting black box function is optimal, so 10 hidden layer nodes are selected; the output layer has 2 output parameters, corresponding to the head and efficiency of the centrifugal pump.

[0088] Initialize weights and thresholds: Use the rand function in the programming software to randomly initialize W1 and b1.

[0089] Select activation function: The activation function of the neural network in the present invention adopts the Sigmoid function, which is a nonlinear function that allows the neural network to learn nonlinear relationships. Multiple neurons in the neural network can represent complex nonlinear functions through the combination of Sigmoid functions. It has a smooth gradient in most of its input range, which is important for the stability and convergence of the gradient descent optimization algorithm. Therefore, the sigmoid function is selected as the activation function of the neural network, and the output range is mapped to between 0 and 1. Its function form is:

[0090]

[0091] Forward propagation: This design passes the input data to the network and calculates the output of each neuron through the neurons in each layer. The BP neural network propagation process is as follows: Figure 2 shown.

[0092] Among them, the input layer is represented by L1, the output layer is represented by L2, the weight between L1 and the hidden layer is represented by V, and the matrix size is 50*5; the weight between L2 and the hidden layer is represented by W, and the matrix size is 2*50.

[0093] Backpropagation: Update the weights and biases by calculating the gradient of the loss function with respect to these parameters to reduce the value of the loss function. Backpropagation uses the gradient descent algorithm or its variants to adjust the parameters to minimize the loss.

[0094] Step 2: Programming the artificial bee colony algorithm to optimize the centrifugal pump output parameters

[0095] Based on the artificial bee colony algorithm, the three roles of leading bees, following bees, and scout bees are used to generate new design parameter combinations, and the performance of each nectar source is evaluated. The optimal solution is determined by comparing the amount of nectar in each place. After multiple iterations of the artificial bee colony algorithm, the optimal food solution is continuously updated. It is characterized by searching for the best design parameters of nearby nectar sources through this cyclic optimization process, including but not limited to the impeller geometry, the inlet and outlet placement angles of the front and rear cover blades, etc., so as to maximize or minimize multiple performance and constraint targets. The specific process is as follows:

[0096] Define the fitness function fit: In the multi-objective optimization process, the centrifugal pump head is required to meet the given head requirement while improving the centrifugal pump efficiency and optimizing the input structural parameters, that is, the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle And to make the centrifugal pump most efficient, the output results are screened through greedy comparison of conditional statements in the fitness function, and the mathematical model of the theoretical head of the centrifugal pump is imported for calculation. When the head reaches the requirement, the efficiency of the centrifugal pump is output and the input parameters corresponding to the maximum efficiency are screened as the optimal solution;

[0097] Artificial bee colony algorithm parameter setting: Establish a zero matrix to store the initial honey source. The so-called initial honey source, in this invention, is to store the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle Column vector of the five input parameters.

[0098] In the present invention, the number of nectar sources is set to 100, the number of leading bees is set to 100, the number of iterations is set to 200, the acceleration coefficient is set to 1, and the random number is constrained between the set minimum and maximum structural parameter values ​​using the following formula in the randomly found nectar source and the optimal solution is searched in this interval:

[0099] zzza=unifrnd(varmin1,varmax1);

[0100] zzzb=unifrnd(varmin2,varmax2);

[0101] zzzc=unifrnd(varmin3,varmax3);

[0102] zzzd=unifrnd(varmin4.varmax4)

[0103] zzze=unifrnd(varmin5,varmax5);

[0104] pop(i).position=[zzza zzzb zzzc zzzd zzze];

[0105] pop(i).costt=sim(net,pop(i).position′);

[0106] Artificial bee colony algorithm programming: The artificial bee colony algorithm iteration includes four stages: initialization, discovery, information exchange, and abandonment.

[0107] The initialization process is to randomly generate a certain number of initial bee individuals, each of which represents a potential solution.

[0108] The discovery process is that the leading bee randomly searches for a new nectar source near nectar source i. When the fitness of the newly found high-quality nectar source is greater than that of the old nectar source, the high-quality nectar source will be selected to replace the original nectar source.

[0109] The information exchange process adopts the roulette method, which puts the structural parameter combination of each nectar source into the fitness function for calculation, and returns the calculated lift and efficiency. The probability of a follower bee selecting a leading bee in the roulette is the ratio of the efficiency return value of the nectar source to the sum of the efficiency return values ​​of all nectar sources.

[0110] The abandonment process is that when the trial iterative search approaches the threshold L in the search process, if the nectar source has not been updated to a better nectar source, the nectar source will be abandoned, and the scout bee will jump out of the nearby nectar source range and find a new nectar source.

[0111] Iteration: Repeat the iterative process. Since the fitness function contains the BP neural network to predict the efficiency of each nectar source, when running the artificial bee colony algorithm, it is necessary to first run the BP neural network to establish a prediction function for the input and output parameters. After iteration, the optimal solution is obtained.

[0112] Step 3: Obtain the final population through the artificial bee colony algorithm, the purpose is to find the optimal solution within the range of optimization parameter variables.

[0113] After multiple iterations, the best fitness function in each iteration is selected as the output. When the output fitness function converges, it is the optimal solution.

[0114] The corresponding example of the present invention needs to optimize the efficiency according to the initial centrifugal pump parameters, where the required head needs to meet the head of 1685m, β s1 =48.9°, β hi =60.8°, β s2 =12.4°, β h2 =14.0°, is the structural parameter to be optimized, and its efficiency is 75.90%. After optimization, the lift meets the requirements and the efficiency is maximized. After consulting the data, the optimization range of the input parameters of the artificial bee colony algorithm is: β s1 =46~51,β h1 =57~62,β s2 =11~14,β h2 =14.5~17.5, Optimize the input parameters within this range. The following is the specific operation process:

[0115] First run the BP neural network in the programming software, and the regression diagram of the training model is as follows Figure 3 The R values ​​of the training set, validation set, and test set are 0.99961, 0.99855, and 0.99956, respectively, and the total R value is 0.99942. It can be considered that this black box function has high reliability.

[0116] Run the artificial bee colony algorithm to perform iterative processes such as initialization, discovery, information exchange, and abandonment, such as Figure 4 , the running results are as follows Figure 5 The attached figure shows that the head and efficiency after optimization are 1712.71m and 76.72% respectively.

[0117] After multiple iterations, the fitness function tends to be stable, and the efficiency at this time can be considered to be the optimal value and the optimized centrifugal pump parameters are: β s1 =57.2°, β h1 =46°, β s2=15.7°, β h2 =14.0°,

[0118] Step 4: Carry out 3D modeling of the input parameters optimized by the artificial bee colony algorithm and perform CAE simulation analysis through relevant software, and compare the simulation prediction results with the output results of the neural network;

[0119] 3D modeling: 3D modeling is performed using relevant software, mainly modeling the fluid domain of the impeller and volute, such as Figure 6 shown.

[0120] Simulation analysis: The simulation analysis is carried out through relevant software. It mainly simulates the inlet and outlet pressures and calculates the lift. The formula is as follows;

[0121]

[0122] Where P out is the centrifugal pump outlet pressure, P in is the inlet pressure of the centrifugal pump, and γ is the bulk density of the conveying medium.

[0123] Simulate the shaft power of the 3D model and calculate the efficiency of the centrifugal pump. The formula is as follows:

[0124]

[0125] Where P is the shaft power of the centrifugal pump and Q is the design flow rate of the centrifugal pump.

[0126] The pressure cloud diagram of the centrifugal pump after simulation is as follows Figure 7 As shown, the final optimized centrifugal pump efficiency is 76.72% and the head is 1712.71m.

[0127] Compared with the centrifugal pump before optimization, the efficiency increased from the initial 75.90% to 76.72%, and the head increased from 1685.04m before optimization to 1712.71m, meeting the head condition: H>1680m. Therefore, it can be considered that the optimization scheme of the present invention can improve the optimization efficiency, reduce the time cost and realize multi-objective optimization design for the design optimization of aviation fuel centrifugal pump by using intelligent algorithms and optimization technology.

[0128] Further, the principles and features of the present invention are described in conjunction with a centrifugal pump example and accompanying drawings. The overall process is as follows: Figure 1 As shown, the examples are only used to explain the present invention, and are not used to limit the scope of the present invention.

[0129] Among the centrifugal pump parameters, the structural parameters of the impeller outlet side have a great influence on the performance of the centrifugal pump, which can be deduced according to the Euler formula:

[0130] gH=(u2V u2 -u1Vu1 )η h

[0131] Where H is the centrifugal pump head, u1 and u2 are the inlet and outlet circumferential speeds of the centrifugal pump impeller, V u1 、V u2 Represents the circular component of the absolute velocity at the inlet and outlet, η h Represents the hydraulic efficiency of the centrifugal pump.

[0132] From the exit velocity triangle:

[0133]

[0134] Where V m2 represents the axial velocity at the outlet, so:

[0135]

[0136] If the inlet of the aviation centrifugal pump is considered to be an illegal inlet, the following formula is obtained:

[0137]

[0138] Where V m1 Indicates the axial speed at the inlet.

[0139] It can be seen in the Euler formula that if the centrifugal pump has an illegal inlet, that is, V u1 ≠0, the lift is closely related to β1 and β2.

[0140] Qualitative analysis of the wrap angle shows that a larger wrap angle is usually associated with a higher head because it can increase the centrifugal force of the pump, thereby increasing the head. However, an excessively large wrap angle will cause the flow channel to become longer, increase friction loss, and reduce the performance of the centrifugal pump. Therefore, it is difficult to find a suitable wrap angle for different pumps to achieve optimal pump performance.

[0141] Therefore, the present invention uses an artificial bee colony algorithm to calculate the structural parameter of the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle Multi-objective optimization of lift and efficiency is carried out, and intelligent algorithms and optimization technologies are used to reduce time costs and improve optimization efficiency for the design optimization of aviation fuel centrifugal pumps.

[0142] Select a centrifugal pump parameter for optimization:

[0143] D2=101mm, D1=42mm, b2=4mm, β s1 =48.°, βhl =60.8°, βs2=12.4°,

[0144]

[0145] Constrained head H>1685m, structural parameter β s1 , β s2 , β h1 , β h2 , In the present invention, 50 groups of sample points are taken as training sets and 30 groups of sample points are taken as prediction sets to perform multi-objective optimization of centrifugal pumps.

[0146] Compared with the centrifugal pump before optimization, the efficiency increased from the initial 75.90% to 76.72%, and the head increased from 1685.04m before optimization to 1712.71m. Therefore, it can be considered that the optimization scheme of the present invention can improve the optimization efficiency, reduce the time cost and realize multi-objective optimization design for the design optimization of aviation fuel centrifugal pump by using intelligent algorithms and optimization technology.

[0147] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A multi-objective optimization method for aviation fuel centrifugal pump based on artificial bee colony algorithm, characterized in that: The steps include: Step 1: Build a BP neural network; Step 2: Programming the artificial bee colony algorithm and using the BP neural network described in step 1 as a black box function to optimize the output parameters of the centrifugal pump; Step 3: Obtain the final nectar source through the artificial bee colony algorithm programming; Step 4: Carry out 3D modeling of the input parameters (i.e., nectar source) after optimization by the artificial bee colony algorithm and perform CAE simulation analysis through relevant software. Compare the simulation prediction results with the output results of the neural network, and compare their head errors and efficiency errors. Then compare the optimized results with the results before optimization, and compare their head and efficiency improvement values.

2. The multi-objective optimization method for aviation fuel centrifugal pump based on artificial bee colony algorithm according to claim 1 is characterized in that: The step 1 is specifically as follows: A BP neural network is established based on existing centrifugal pump samples, and the prediction accuracy is improved through experimental data training and verification to adapt to fuel centrifugal pumps of different types and sizes, and accurately predict the performance parameters of fuel centrifugal pumps at multiple operating points, including but not limited to head and efficiency.

3. The multi-objective optimization method for aviation fuel centrifugal pump based on artificial bee colony algorithm according to claim 2 is characterized in that: For the parameter optimization of aviation centrifugal pumps, the construction of BP neural network specifically includes the following processes: (1) Data preparation: Prepare data sets for training and testing neural networks; collect data on aviation centrifugal pumps; The input parameters, namely the BP neural network input layer, include the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle The output parameters, i.e. the output layer of the BP neural network, include the centrifugal pump head H and efficiency η. The data is divided into a training set, a prediction set, and a validation set, and the corresponding input layer and output layer matrix sizes are (50*5) and (2*50), respectively. (2) Determine the network architecture: The constructed BP neural network includes an input layer, a hidden layer and an output layer. The input layer has 5 parameters. After comparing the network prediction results of different numbers of nodes in the hidden layer, it is finally determined that when the hidden layer nodes are 10, the fitting black box function is optimal, so 10 hidden layer nodes are selected. The output layer has 2 output parameters, corresponding to the centrifugal pump head and efficiency. (3) Initialize weights and thresholds: Use the rand function in the programming software to randomly initialize W1 and b1; Where: W1 is the weight, b1 is the bias; (4) Select activation function: The activation function of the neural network uses the Sigmoid function and maps the output range to between 0 and 1. Its function form is: (4) Forward propagation: pass the input data to the network and calculate the output of each neuron through the neurons in each layer; The input layer is represented by L1, the output layer is represented by L2, the weight between L1 and the hidden layer is represented by V, and the matrix size is 50*5; the weight between L2 and the hidden layer is represented by W, and the matrix size is 2*50; (5) Backpropagation: Update the parameters by calculating the gradient of the loss function with respect to the weights and biases to reduce the value of the loss function. Backpropagation uses the gradient descent algorithm or its variants to adjust the parameters to minimize the loss.

4. The multi-objective optimization method for aviation fuel centrifugal pump based on artificial bee colony algorithm according to claim 1 is characterized in that: The step 2 is specifically as follows: based on the artificial bee colony algorithm, the three roles of the leading bee, the following bee and the scout bee are used to generate a new combination of design parameters, and the performance of each nectar source is evaluated, and the optimal solution is determined by comparing the amount of nectar at each location. After multiple iterations of the artificial bee colony algorithm, the optimal solution of food is continuously updated; This cyclic optimization process is used to search for the best design parameters of the nearby honey source, including but not limited to the impeller geometry, the inlet and outlet placement angles of the front and rear cover blades, so as to maximize or minimize multiple performance and constraint objectives.

5. The multi-objective optimization method for aviation fuel centrifugal pump based on artificial bee colony algorithm according to claim 4 is characterized in that: The specific operation of designing parameters in step 2 is: S1 defines the fitness function fit: In the multi-objective optimization process, the centrifugal pump head is required to meet the given head requirement while improving the centrifugal pump efficiency and optimizing the input structural parameters, that is, the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle And to make the centrifugal pump most efficient, the output results are screened through greedy comparison of conditional statements in the fitness function, and the mathematical model of the theoretical head of the centrifugal pump is imported for calculation. When the head reaches the requirement, the efficiency of the centrifugal pump is output and the input parameters corresponding to the maximum efficiency are screened as the optimal solution; S2 Artificial bee colony algorithm parameter setting: establish a zero matrix to store the initial nectar source array; The so-called initial honey source is the centrifugal pump impeller front cover blade inlet placement angle β s1 、Front cover blade outlet placement angle β s2 、The rear cover blade inlet placement angle β h1 、Rear cover blade outlet placement angle β h2 And the blade angle Column vector of the five input parameters. In the randomly found honey source, the following formula is used to constrain the random number between the minimum and maximum structural parameter values ​​and search for the optimal solution in this interval: zzza=unifrnd(varmin1,varmax1); zzzb=unifrnd(varmin2,vatmax2); zzzc=unifrnd(varmin3,varmax3); zzzd=unifrnd(varmin4,varmax4); zzze=unifrnd(varmin5,varmax5); pop(i).position=[zzza zzzb zzzczzzd zzze]; pop(i).cost=sim(net, pop(i).position′); The first five rows represent the continuous and uniform generation of random numbers within the range of optimized parameter values; The 6th row represents placing the randomly generated parameters in a space; Line 7 represents substituting the generated parameters into the black box function to calculate the result; S3 Artificial Bee Colony Algorithm Programming: Artificial Bee Colony Algorithm iteration includes initialization, discovery, information exchange, and abandonment; S4 iteration: Repeat the iteration process. Since the fitness function contains the bp neural network to predict the efficiency of each nectar source, when running the artificial bee colony algorithm, it is necessary to first run the bp neural network to establish a prediction function for the input and output parameters. After iteration, the optimal solution is obtained.

6. The multi-objective optimization method for aviation fuel centrifugal pump based on artificial bee colony algorithm according to claim 5, characterized in that: The initialization process in S3 is to randomly generate a certain number of initial bee individuals, each bee represents a potential solution; The discovery process in S3 is that the leading bee randomly searches for a new nectar source near nectar source i. When the fitness of the newly found high-quality nectar source is greater than that of the old nectar source, the high-quality nectar source will be selected to replace the original nectar source. The information exchange process in S3 adopts the roulette method, puts the structural parameter combination of each nectar source into the fitness function for calculation, and returns the calculated lift and efficiency. The probability of a follower bee selecting a leading bee in the roulette is the ratio of the efficiency return value of the nectar source to the sum of the efficiency return values ​​of all nectar sources; The abandonment process in S3 is that when the trial iterative search approaches the threshold L in the search process, if the nectar source has not been updated to a better nectar source, the nectar source will be abandoned, and the scout bee will jump out of the nearby nectar source range and find a new nectar source.

7. The multi-objective optimization method for aviation fuel centrifugal pump based on artificial bee colony algorithm according to claim 1 is characterized in that: The step three is specifically as follows: after multiple iterations of nectar source updates, the best fitness function in each iteration process is selected as the output, and when the output fitness function converges, it is the optimal solution; The efficiency is optimized according to the initial centrifugal pump parameters, where the required head must meet 1685m head, β sl =48.9°, β hl =60.8°, β s2 =12.4°, β h2 =14.0°, The structural parameters to be optimized have an efficiency of 75.90%. After optimization, the lift meets the requirements and the efficiency is maximized. The optimization range of the input parameters of the artificial bee colony algorithm is: β s1 =46~51,β h1 =57~62,β s2 =11~14,β h2 =14.5~17.5, Optimize the input parameters within this range. The following is the specific operation process: First run the BP neural network in the programming software, and its training model regression is as follows: Run the artificial bee colony algorithm to perform iterative processes such as initialization, discovery, information exchange, and abandonment; After multiple iterations, the fitness function tends to be stable, and the efficiency at this time can be considered to be the optimal value and the optimized centrifugal pump parameters are: β sl =57.2°, β h1 =46°, β s2 =15.7°, β h2 =14.0°, 8. The multi-objective optimization method for aviation fuel centrifugal pump based on artificial bee colony algorithm according to claim 1 is characterized in that: The step 4 specifically includes: performing three-dimensional modeling on the input parameters optimized by the artificial bee colony algorithm and performing CAE simulation analysis through relevant software, and comparing the simulation prediction results with the output results of the neural network; S1 3D modeling: 3D modeling is performed using relevant software, mainly modeling the fluid domain of the impeller and volute; S2 simulation analysis: The simulation analysis is carried out through relevant software, mainly simulating the inlet and outlet pressures and calculating the lift. The formula is as follows; Where P out is the centrifugal pump outlet pressure, P in is the inlet pressure of the centrifugal pump, and γ is the bulk density of the conveying medium. Simulate the shaft power of the 3D model and calculate the efficiency of the centrifugal pump. The formula is as follows: Where P is the shaft power of the centrifugal pump and Q is the design flow rate of the centrifugal pump.

Images