Aerodynamic parameter identification method based on adaptive genetic algorithm and neural network

Abstract

This application discloses a method for identifying aerodynamic parameters based on an adaptive genetic algorithm and a neural network. The method includes: establishing a neural network model based on the acquired angle of attack, flight Mach number, elevator angle, and aerodynamic coefficients of a flying object; optimizing the initial weights and thresholds of the neural network model using an adaptively adjusted genetic algorithm; and training the neural network model based on the optimized initial weights and thresholds; identifying the real-time aerodynamic coefficients of the flying object using the optimized and trained neural network model, and identifying the real-time aerodynamic derivatives of the flying object using the derivative characteristics of the neural network model. This invention solves the technical problem of inaccurate aerodynamic parameter identification results in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an aerodynamic parameter identification method based on an adaptive genetic algorithm and a neural network. Background Art

[0002] Traditional aerodynamic identification involves measuring missile input and output data during testing, building a mathematical model that reflects the missile's nonlinear characteristics, and identifying the unknown coefficients within the model. Common methods for obtaining aerodynamic parameters include wind tunnel testing, theoretical calculations, flight testing, and establishing an aerodynamic identification model using the maximum likelihood criterion. This transforms the aerodynamic parameter identification problem into an optimization problem, optimizing the model parameters to minimize the deviation between the model output and the measured values.

[0003] However, traditional methods often struggle to obtain experimental data, and the modeling process is complex. Neural networks, with their strong nonlinear mapping and function approximation capabilities, have been increasingly applied to aerodynamic parameter identification in recent years. Yuan Zhijie constructed a BP neural network model using a mind evolution algorithm to predict missile aerodynamic parameters. Carpenter M et al. used neural networks to implement a multivariable function approximation method for missile aerodynamic parameter prediction. Pu Jialun et al. proposed a method for sample augmentation and online rapid correction of neural network parameters based on support vector machines (SVMs), demonstrating the feasibility of SVM-based neural network methods for aircraft aerodynamic parameter identification. In engineering applications or ballistic simulation, the missile's flight state is often calculated by interpolating the corresponding aerodynamic parameters into the missile's equations of motion based on the Mach number, angle of attack, and rudder deflection. However, this method is significantly affected by the interpolation points and has a relatively small solution range.

[0004] The disadvantages of the prior art are:

[0005] 1) The modeling process is relatively complicated;

[0006] 2) The initial parameters of the neural network are randomly selected, and the identification results are prone to fall into local optimality;

[0007] 3) The aerodynamic coefficients and aerodynamic derivatives need to be modeled and identified separately, and the identification results need to be interpolated into the missile motion equations;

[0008] 4) Interpolation is highly dependent on data points (to ensure accuracy, a sufficient number of interpolation points are required), and is constrained by the interpolation points, so its solution range is relatively small and it does not have predictive capabilities.

[0009] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0010] The embodiment of the present invention provides an aerodynamic parameter identification method based on an adaptive genetic algorithm and a neural network, so as to at least solve the technical problem of inaccurate aerodynamic parameter identification results in the prior art.

[0011] According to one aspect of an embodiment of the present invention, a method for identifying aerodynamic parameters based on an adaptive genetic algorithm and a neural network is provided, comprising: establishing a neural network model based on the acquired angle of attack, flight Mach number, elevator deflection angle, and aerodynamic coefficient of a flying body; optimizing the initial weights and thresholds of the neural network model using an adaptively adjusted genetic algorithm, and training the neural network model based on the optimized initial weights and thresholds; identifying the real-time aerodynamic coefficients of the flying body using the optimized and trained neural network model, and identifying the real-time aerodynamic derivatives of the flying body using the derivative characteristics of the neural network model, wherein the aerodynamic parameters include the aerodynamic coefficients and the aerodynamic derivatives.

[0012] According to another aspect of an embodiment of the present invention, an aerodynamic parameter identification device based on an adaptive genetic algorithm and a neural network is provided, comprising: an establishment module configured to establish a neural network model based on the acquired angle of attack, flight Mach number, elevator deflection angle, and aerodynamic coefficient of the flying body; an optimization module configured to optimize the initial weights and thresholds of the neural network model using an adaptively adjusted genetic algorithm, and train the neural network model based on the optimized initial weights and thresholds; an identification module configured to identify the real-time aerodynamic coefficients of the flying body using the optimized and trained neural network model, and identify the real-time aerodynamic derivatives of the flying body using the derivative characteristics of the neural network model; wherein the aerodynamic parameters include the aerodynamic coefficients and the aerodynamic derivatives.

[0013] In an embodiment of the present invention, an identification model is established using existing data, with angle of attack, Mach number, and rudder angle as inputs and related aerodynamic coefficients as outputs; a genetic algorithm with adaptive adjustment of crossover and mutation probabilities is used to optimize the initial weights and thresholds of the neural network, effectively preventing the identification results from falling into local optimality; the derivative characteristics of the neural network are used to further identify the aerodynamic derivatives of the corresponding aerodynamic coefficients; the identification results can be directly introduced into the missile motion equations for trajectory calculation without interpolation, thereby realizing the integrated identification and application of aerodynamic parameters, and thus solving the technical problem of inaccurate aerodynamic parameter identification results in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0015] Figure 1is a flow chart of an aerodynamic parameter identification method based on an adaptive genetic algorithm and a neural network according to a first embodiment of the present invention;

[0016] Figure 2 is a flow chart of an aerodynamic parameter identification method based on an adaptive genetic algorithm and a neural network according to a second embodiment of the present invention;

[0017] Figure 3 is a neural network topology diagram according to an embodiment of the present invention;

[0018] Figure 4 is a flow chart of an aerodynamic parameter identification method based on an adaptive genetic algorithm and a neural network according to a third embodiment of the present invention;

[0019] Figure 5 is a flow chart of an aerodynamic parameter identification method based on an adaptive genetic algorithm and a neural network according to a fourth embodiment of the present invention;

[0020] Figure 6 is a comparison diagram of aerodynamic coefficient identification curves according to an embodiment of the present invention;

[0021] Figure 7 is a comparison diagram of aerodynamic derivative identification curves according to an embodiment of the present invention;

[0022] Figure 8 is a comparison diagram of aerodynamic derivative identification curves according to an embodiment of the present invention;

[0023] Figure 9 is a speed comparison curve diagram according to an embodiment of the present invention;

[0024] Figure 10 is a flight trajectory comparison curve diagram according to an embodiment of the present invention.

[0025] Overview

[0026] The present invention combines an adaptive genetic algorithm with a neural network and applies it to the aerodynamic identification problem of micro-guided munitions, and proposes an adaptive genetic algorithm-back propagation neural network (AGA-BPNN) aerodynamic identification model.

[0027] Using existing data, an identification model is established with angle of attack, Mach number, and rudder deflection as inputs and the relevant aerodynamic coefficients as outputs. A genetic algorithm with adaptive crossover and mutation probabilities is used to optimize the initial weights and thresholds of the neural network, effectively preventing the identification results from falling into local optima. The derivative properties of the neural network are then used to further identify the aerodynamic derivatives of the corresponding aerodynamic coefficients. The identification results can be directly incorporated into the missile's equations of motion for trajectory calculation without interpolation, achieving integrated aerodynamic parameter identification and application. This method can also be applied to ammunition aerodynamic layout analysis and control system design. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described 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 making creative efforts should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] Example 1

[0031] According to an embodiment of the present invention, a method for identifying aerodynamic parameters based on an adaptive genetic algorithm and a neural network is provided. Figure 1 As shown, the method includes:

[0032] Step S102: establishing a neural network model based on the acquired angle of attack, flight Mach number, elevator angle, and aerodynamic coefficient of the flying object.

[0033] For example, a nonlinear identification model is established based on the angle of attack, the flight Mach number, the elevator deflection angle, and the aerodynamic coefficient; and a neural network is used to fit the nonlinear identification model to establish the neural network model.

[0034] In an exemplary embodiment, the following method can be specifically used to establish the neural network model: the angle of attack, the flight Mach number, and the elevator deflection angle are used as the three neurons of the input layer of the neural network; the aerodynamic coefficients are used as the neurons of the output layer of the neural network, wherein the aerodynamic coefficients include the drag coefficient, lift coefficient, pitch moment coefficient and the relative position coefficient of the pressure center of the flying body to the center of mass of the flying body; based on the settings of the neurons of the input layer and the output layer of the neural network, the neural network model is trained so that the neural network model fits the nonlinear identification model to establish the neural network model.

[0035] Step S104 , optimizing the initial weights and thresholds of the neural network model using a genetic algorithm with adaptive adjustment, and training the neural network model based on the optimized initial weights and thresholds.

[0036] In one exemplary embodiment, the initial weights and thresholds of the neural network model are encoded to generate an initial population, which is then used as the current population to be optimized. For example, a real number encoding scheme is used to encode the initial weights and thresholds of the neural network model to form chromosome vectors of individuals in the initial population, thereby determining the initial population. The initial population includes multiple individuals, and the chromosome vector of each individual includes all weights and thresholds of the neural network model.

[0037] Then loop through the following steps until the number of iterations reaches the preset total number of iterations:

[0038] Calculating the fitness value of the individuals in the population to be optimized, and determining whether the fitness value meets the optimization goal;

[0039] If the fitness value meets the optimization goal, the initial weights and thresholds of the optimized neural network model are obtained, and the loop is exited. Otherwise, the population to be optimized is selected, cross-mutated, and a new generation of population is generated. The new generation of population is used as the population to be optimized, and the number of iterations is increased by 1.

[0040] In an exemplary embodiment, the specific process of selection, crossover, and mutation can be as follows: determine the probability of the population individual being selected based on the fitness value of the individual in the population to be optimized, and select the target individual from the population to be optimized; adaptively adjust the crossover probability, and use the adaptively adjusted crossover probability to perform a crossover operation on the chromosome vector of the target individual; adaptively adjust the mutation probability, and use the adaptively adjusted mutation probability to perform a mutation operation on the chromosome vector of the individual in the population to be optimized after the crossover operation to generate a new generation of population.

[0041] In an exemplary embodiment, the adaptive adjustment of the crossover probability can be: increasing the crossover probability of the population with poor average fitness value in the next iteration during each iteration, reducing the crossover probability of the population with good average fitness value in the next iteration, and gradually reducing the overall crossover probability as the number of iterations increases.

[0042] In an exemplary embodiment, the adaptive adjustment of the crossover probability can also be: adaptively adjusting the crossover probability based on the maximum crossover probability, the current number of iterations, the total number of iterations, the fitness value of the population individual with the smallest fitness value in the current iteration, and the average fitness value of the population in the current iteration.

[0043] In an exemplary embodiment, adaptive adjustment of the mutation probability can be: increasing the mutation probability of the population with poor average fitness value in each iteration in the next iteration, reducing the mutation probability of the population with good average fitness value in the next iteration, and gradually reducing the overall mutation probability as the number of iterations increases.

[0044] In an exemplary embodiment, the adaptive adjustment of the mutation probability can also be: adaptively adjusting the mutation probability based on the maximum mutation probability, the current number of iterations, the total number of iterations, the fitness value of the population individual with the smallest fitness value in the current iteration, and the average fitness value of the population in the current iteration.

[0045] Step S106 , using the optimized trained neural network model to identify the real-time aerodynamic coefficients of the flying body, and using the derivative characteristics of the neural network model to identify the real-time aerodynamic derivatives of the flying body.

[0046] In an exemplary embodiment, using the optimized trained neural network model to identify the real-time aerodynamic coefficients of the flying body includes: obtaining the real-time angle of attack, flight Mach number, and elevator angle of the flying body, and inputting the real-time angle of attack, flight Mach number, and elevator angle as inputs into the optimized trained neural network model to obtain the real-time aerodynamic coefficients of the flying body.

[0047] In an exemplary embodiment, using the derivative characteristics of the neural network model to identify the real-time aerodynamic derivatives of the flying body includes: using the neural network model to obtain the real-time aerodynamic coefficient, and then using the derivative characteristics of the neural network to obtain the aerodynamic derivatives of the real-time aerodynamic coefficient with respect to the real-time angle of attack, flight Mach number, and elevator angle of the flying body.

[0048] The aerodynamic parameters include the aerodynamic coefficient and the aerodynamic derivative.

[0049] In this embodiment, an improved adaptive genetic algorithm is used to optimize the initial weights and thresholds of the neural network, which effectively avoids the identification results falling into local optimality due to random selection of the initial weights and thresholds of the neural network, thereby improving the identification accuracy.

[0050] Example 2

[0051] According to an embodiment of the present invention, another aerodynamic parameter identification method based on adaptive genetic algorithm and neural network is provided. Figure 2 As shown, the method includes:

[0052] Step S202: system modeling.

[0053] For axisymmetric missiles, ignoring the influence of factors such as altitude, rotation angular rate, angle of attack change rate, and rudder angle change rate, the aerodynamic coefficient can be regarded as a nonlinear function of angle of attack, rudder angle, and Mach number. The nonlinear identification model can be expressed as follows:

[0054]

[0055] Among them, c D is the drag coefficient, c L is the lift coefficient, c z is the lateral force coefficient, m z is the pitching moment coefficient, X CP is the relative position coefficient of the missile's pressure center compared to its mass center, α is the angle of attack, and δ z is the elevator deflection angle, Ma is the flight Mach number, f1, f2, f3, and f4 are the first, second, third, and fourth fitting functions respectively.

[0056] Step S204: construct a neural network model.

[0057] For general nonlinear mapping problems, a three-layer neural network can achieve good mapping effects. The neural network topology used in the embodiment of the present invention is as follows: Figure 3 As shown, the input layer has three neurons, corresponding to the angle of attack, rudder angle, and Mach number. The hidden layer has six neurons, and the output layer has four neurons, corresponding to the aerodynamic coefficient c D 、c L 、m z 、X CP .

[0058] The main purpose of the neural network is to use the output vector z to approximate the true value vector z*. The output error term of the network is defined as:

[0059] e c =zz* (2)

[0060] Thus, the network objective function can be obtained as:

[0061]

[0062] Among them, n is the dimension of the output layer vector, m is the number of dataset vector groups, a represents the ath vector of the output vector set and the true value vector set, c is the cth element of the current output vector and the current true value vector, z ac is the cth element of the ath vector of the current output vector set, z ac is the cth element of the ath vector of the true value vector set.

[0063] Set the input layer vector to x and the hidden layer vector to y, then the network input layer to the output layer:

[0064] z=g(v T (f(w T x+b1))+b2) (4)

[0065] Where b1 is the threshold vector from the input layer to the hidden layer, b2 is the threshold vector from the hidden layer to the output layer, f(·) is the nonlinear function used by the hidden layer, and g(·) is the activation function of the output layer. w is the weight matrix from the network input layer to the hidden layer, and v is the weight matrix from the hidden layer to the output layer, where w jk represents the weight from the jth neuron in the input layer to the kth neuron in the hidden layer, v ij represents the weight from the kth neuron in the hidden layer to the ith neuron in the output layer, T represents the transpose of the vector, and z represents the output vector.

[0066] The activation function g(·) of the output layer is a linear function, and the activation function f(·) of the hidden layer network is expressed as:

[0067]

[0068] Here, x represents a process quantity, which is the value of the hidden layer after the data is imported from the input layer, and e is a mathematical value representing a natural constant.

[0069] Using the chain differential rule and the principle of gradient descent, we can get the correction formula for network weights and thresholds:

[0070]

[0071] Among them, k is the number of iterations, η is the learning rate, x represents the input layer, y represents the hidden layer, and z represents the output layer. At this point, the back propagation process of the neural network has been completed.

[0072] Step S206: Optimize the neural network model using an adaptive genetic algorithm.

[0073] To address the problem that random selection of initial weights and thresholds in neural networks easily leads to local optima, the present invention introduces a genetic algorithm to optimize the neural network model. To increase the global search capability and convergence speed of the genetic algorithm, the crossover and mutation probabilities of the traditional genetic algorithm are adaptively adjusted.

[0074] In an exemplary embodiment, the specific optimization method may include the following steps:

[0075] (1) Population initialization operation

[0076] The real number encoding method is used for encoding. The chromosome vector of the population individual includes all the weights and thresholds of the neural network. The specific arrangement is:

[0077] X i =[w 11 ...w 36 , b 11 ...b 16 , v 11 ...v 64 , b 21 ...b 24 ] (7)

[0078] Among them, X i represents the chromosome vector of the i-th individual in the population, w 11 …w 36 Represents the weights of each neuron in the input layer to each neuron in the hidden layer, b 11 …b 16 Respectively represent the thresholds of each neuron from the output layer to the hidden layer, v 11 …v 64 Represents the weights of each neuron in the hidden layer to each neuron in the output layer, b 21 ...b 24 They represent the thresholds of each neuron from the hidden layer to the output layer.

[0079] (2) Select operation

[0080] Using the objective function E c The value is the fitness value v of the individual in the population i , the probability of an individual being selected is:

[0081]

[0082] Among them, ζ is the coefficient and N is the population size.

[0083] (3) Crossover operation

[0084] Abandoning the traditional method of determining the value based on crossover probability, this method adaptively adjusts the crossover probability using a functional relationship. This increases the crossover probability of populations with poor average fitness values ​​in the next iteration, while decreasing the crossover probability of populations with good average fitness values ​​in the next iteration. To ensure both early global search capability and later convergence, the overall crossover probability is gradually reduced as the number of iterations increases.

[0085] The calculation formula for the crossover probability is as follows:

[0086]

[0087] Among them, P c is the crossover probability of the next generation of the population, P c max is the maximum crossover probability set, P cx For a value greater than 0 and less than P c maxThe number of iterations, n is the current number of iterations, N is the total number of iterations, F best is the fitness value of the population individual with the smallest fitness value in the current iteration, F agvn is the average fitness value of the population in the current iteration, and μ is the influencing factor. The formula for chromosome crossover is:

[0088]

[0089] Formula (10) represents the chromosome vector a of the i-th individual in the population i The crossover operation at position k with the chromosome vector aj of the jth individual in the population at position k, b is a random number between 0 and 1, a ik Represents the chromosome vector a of the i-th individual in the population i The gene value at position k, a jk Represents the gene value at position k of the chromosome vector aj of the jth individual in the population.

[0090] (4) Mutation operation.

[0091] In order to further improve the global search capability of the algorithm, the adaptive calculation formula of the mutation probability is the same as the crossover operation:

[0092]

[0093] Among them, P m is the probability of mutation in the next generation of the population, P mmax is the maximum mutation probability, P mx is greater than 0 and less than P mmax The formula for the variation method is as follows:

[0094]

[0095]

[0096] Formula (12) means that the gene value of the chromosome vector of the i-th individual in the population is selected for mutation at position j. In formula (12-13), a max is the upper bound of the gene value of the chromosome vector of the population individual, a min is the lower bound of the gene value of the chromosome vector of the population individual, r and r2 are random numbers between 0 and 1, n is the current iteration number, N is the total number of iterations, and a ij Represents the chromosome vector a of the i-th individual in the population i The gene value at position j, f(g) represents a constraint function for the size of the gene value variation.

[0097] Step S208: Identify aerodynamic derivatives using the neural network derivative characteristics.

[0098] During the back propagation process of the neural network, the gradient descent method is used to solve the derivative of the output error with respect to the weight and threshold. Similarly, this idea can be used to solve the derivative of the neural network output with respect to the input.

[0099] After the model training is completed, the corresponding aerodynamic coefficients can be obtained by inputting the corresponding rudder deflection angle, angle of attack and Mach number. The partial derivative of the aerodynamic coefficient with respect to the input, i.e. the aerodynamic derivative, can be identified by using the derivative characteristics of the neural network.

[0100] The aerodynamic derivative identification derivation process is given below. The partial derivative of the neural network output layer vector z with respect to the input layer vector x is as follows:

[0101]

[0102] in

[0103]

[0104] So we can get:

[0105]

[0106] At this point, the aerodynamic derivative identification process based on the derivative characteristics of the neural network has been completed.

[0107] In this embodiment, an improved adaptive genetic algorithm is used to optimize the initial weights and thresholds of the neural network, which effectively avoids the identification results falling into local optimality due to random selection of the initial weights and thresholds of the neural network, thereby improving the identification accuracy.

[0108] Furthermore, in this embodiment, the derivative characteristics of the neural network are utilized to further identify the corresponding aerodynamic derivatives based on the identification model.

[0109] In addition, in the present embodiment, aerodynamic parameters can also be identified in real time according to the flight state, without the need for interpolation, and can be directly brought into the missile motion equations. Interpolation is an approximate function that solves a finite number of known data points. It is a kind of approximate solution method. It is relatively dependent on the data points (to ensure accuracy, enough interpolation points are needed). Because it is constrained by the interpolation points, its solution range is relatively small. Neural networks have very strong nonlinear mapping capabilities and function approximation capabilities. They can not only fit the data points well, but also have very strong predictive capabilities and have a wider solution range.

[0110] Example 3

[0111] According to an embodiment of the present invention, another aerodynamic parameter identification method based on adaptive genetic algorithm and neural network is provided. Figure 4 As shown, the method includes:

[0112] Step S401: data preprocessing.

[0113] The acquired data such as the angle of attack, flight Mach number, elevator angle, aerodynamic coefficient, etc. of the flying body are normalized.

[0114] Step S402: Encode the initial weights and thresholds of the neural network model.

[0115] A real number encoding method is selected to encode the initial weights and thresholds of the neural network model.

[0116] Step S403: generating an initial population.

[0117] After encoding, a chromosome vector of the individual in the initial population is formed, and the initial population is determined, wherein the initial population includes multiple individuals, and the chromosome vector of each individual includes all weights and thresholds of the neural network model. The initial population is used as the population to be optimized.

[0118] Step S404: Calculate the fitness value.

[0119] Calculate the fitness value of the individuals in the population to be optimized, and determine whether the fitness value meets the optimization goal.

[0120] Step S405: whether the optimization goal is met.

[0121] If the fitness value satisfies the optimization goal, the loop is exited and step S408 is executed; otherwise, step S406 is executed to select the population to be optimized and perform crossover and mutation processing to generate a new generation of population.

[0122] Step S406: selection, crossover, and mutation.

[0123] The objective function value is used as the crossover value for each individual in the population, and the probability of an individual being selected is calculated. Subsequently, the traditional method of determining the crossover probability is abandoned, and a functional relationship is used to adaptively adjust the crossover probability. During each iteration, the crossover probability of populations with poor average fitness values ​​is increased in the next iteration, while the crossover probability of populations with good average fitness values ​​is decreased in the next iteration. To ensure both early global search capability and later convergence, the overall crossover probability is gradually reduced as the number of iterations increases. Finally, to further improve the algorithm's global search capability, the chromosome vectors of the individuals in the population after crossover are mutated.

[0124] Step S407: Generate a new generation population.

[0125] After performing the selection, crossover, and mutation operations, a new generation of population is generated, and the process jumps back to step S405.

[0126] Step S408: Obtain the optimal initial weight and threshold.

[0127] Step S409: training the neural network model.

[0128] The obtained optimal initial weights and thresholds are used to train the neural network model.

[0129] Step S410: testing and application.

[0130] After training, the real-time angle of attack, flight Mach number, and elevator deflection are fed into the trained neural network model to obtain the corresponding aerodynamic coefficients. The neural network's derivative properties are then used to identify the aerodynamic derivatives of the aerodynamic coefficients with respect to the corresponding elevator deflection, angle of attack, and other parameters.

[0131] Example 4

[0132] The embodiment of the present invention combines an adaptive genetic algorithm and a neural network for aerodynamic parameter identification, proposes an adaptive genetic algorithm-back propagation neural network aerodynamic identification model, namely a neural network model, and uses the neural network model to identify startup parameters.

[0133] According to an embodiment of the present invention, a pneumatic parameter identification and application process is provided, such as Figure 5 As shown, the method includes:

[0134] Step S501, obtaining the real-time angle of attack, rudder angle, and Mach number according to the flight status.

[0135] Step S502, taking the angle of attack, rudder angle, and Mach number as inputs into the trained neural network model;

[0136] The historically acquired angle of attack, rudder deflection angle, and Mach number are used as input, and the corresponding aerodynamic coefficients are used as output to train the neural network model, which can identify the required aerodynamic coefficients.

[0137] The real-time acquired angle of attack, rudder angle, and Mach number are used as input into the trained neural network model.

[0138] Step S503, obtaining corresponding aerodynamic coefficients through calculation using a neural network model;

[0139] After the real-time acquired angle of attack, rudder angle, and Mach number are brought into the trained neural network model as input, the corresponding aerodynamic coefficients can be calculated using the neural network model.

[0140] Step S504, identifying corresponding aerodynamic derivatives using the neural network derivative characteristics;

[0141] On the basis of the neural network model, the corresponding aerodynamic derivatives can be further obtained according to the derivative characteristics of the neural network.

[0142] Step S505: Substitute the required aerodynamic parameters into the missile motion equations.

[0143] In this embodiment, the neural network model can identify the corresponding aerodynamic coefficients and aerodynamic derivatives based on the real-time flight status (angle of attack, rudder deflection, Mach number) without interpolation, and can be directly applied to the missile motion equations to achieve integrated aerodynamic parameter identification and application.

[0144] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0145] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0146] Example 5

[0147] To verify the practicality of the algorithm, a set of data determined by flight Mach number and rudder angle was selected as a test set. Using the angle of attack as the variable, the three algorithm models, neural network, genetic algorithm-neural network, and adaptive genetic algorithm-neural network, were used to identify various aerodynamic parameters. The crossover probability of the GA-BP neural network algorithm was set to 0.3, the mutation probability was set to 0.05, and the remaining parameters were the same as those of the AGA-BP neural network algorithm. Simulation verification was performed using the angle of attack as the horizontal axis and the various aerodynamic parameters as the variables. The parameter designs for the neural network and adaptive genetic algorithm are shown in the following table:

[0148] Table 1 Neural network parameters

[0149]

[0150] Table 2 Adaptive genetic algorithm parameters

[0151]

[0152] The comparison results of the identification curves of various aerodynamic coefficients under the three algorithms are as follows: Figure 6 shown.

[0153] The mean square errors of the simulation results of the three algorithms corresponding to various aerodynamic parameters are shown in Table 3:

[0154] Table 3 Comparison of mean square error of various parameters

[0155]

[0156] The aerodynamic derivatives are identified using the derivative characteristics of neural networks and the finite difference method. and The results are as follows Figure 7 shown.

[0157] The trajectory is verified by uncontrolled flight of micro-guided munitions, and the trained network model is used to identify the aerodynamic parameters required for the trajectory simulation in the engineering involved in this application. The aerodynamic parameters are then substituted into the six-degree-of-freedom trajectory model and compared with the trajectory calculated by interpolation under the theoretical aerodynamic parameters. Simulation conditions: the missile mass is m = 1.5kg, the missile initial velocity V = 30m / s, the missile initial coordinates are [01.50]m, the initial pitch angle is set to θ = 20°, the initial yaw angle is set to ψ = -3°, the speed-increasing engine starts working at 0.2 seconds, and the speed-increasing engine stops working at 0.6 seconds, and the engine thrust P = 1000N. The velocity comparison curve of the ballistic simulation and the flight trajectory comparison curve in the ground coordinate system are given below. Figure 8 and Figure 9 shown.

[0158] Example 6

[0159] According to an embodiment of the present invention, there is also provided an aerodynamic parameter identification device based on an adaptive genetic algorithm and a neural network for implementing the method of the above embodiment, such as Figure 10 As shown, the device includes:

[0160] An establishing module 12 is configured to establish a neural network model based on the acquired angle of attack, flight Mach number, elevator angle, and aerodynamic coefficient of the flying object;

[0161] an optimization module 13 configured to optimize the initial weights and thresholds of the neural network model using an adaptively adjusted genetic algorithm, and train the neural network model based on the optimized initial weights and thresholds;

[0162] The identification module 14 is configured to use the optimized trained neural network model to identify the real-time aerodynamic coefficients of the flying body, and to use the derivative characteristics of the neural network model to identify the real-time aerodynamic derivatives of the flying body; wherein the aerodynamic parameters include the aerodynamic coefficients and the aerodynamic derivatives.

[0163] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments 1 to 5, and this embodiment will not be repeated here.

[0164] Example 7

[0165] The embodiment of the present invention further provides a storage medium. Optionally, in this embodiment, the storage medium can be configured to store program codes for executing the methods in embodiments 1 to 5.

[0166] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.

[0167] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0168] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above-mentioned computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing one or more computer devices (such as personal computers, servers, or network devices) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0169] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0170] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, and can be electrical or other forms.

[0171] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0172] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0173] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for aerodynamic parameter identification based on adaptive genetic algorithm and neural network, characterized in that: include: A neural network model is established based on the acquired attack angle, flight Mach number, elevator deflection angle and aerodynamic coefficient of the flying body; Optimizing initial weights and thresholds of the neural network model using an adaptively adjusted genetic algorithm, and training the neural network model based on the optimized initial weights and thresholds; Using the optimized trained neural network model to identify the real-time aerodynamic coefficients of the flying body, and using the derivative characteristics of the neural network model to identify the real-time aerodynamic derivatives of the flying body; Wherein, the aerodynamic parameters include the aerodynamic coefficient and the aerodynamic derivative; The adaptive adjustment of the crossover probability includes: increasing the crossover probability of a population with a poor average fitness value in the next iteration during each iteration, reducing the crossover probability of a population with a good average fitness value in the next iteration, and gradually reducing the overall crossover probability as the number of iterations increases; and / or the adaptive adjustment of the mutation probability includes: increasing the mutation probability of a population with a poor average fitness value in the next iteration during each iteration, reducing the mutation probability of a population with a good average fitness value in the next iteration, and gradually reducing the overall mutation probability as the number of iterations increases; The calculation formula for the crossover probability is as follows: Among them, P c is the crossover probability of the next generation of the population, P c max is the maximum crossover probability set, P cx For a value greater than 0 and less than P c max The number of iterations, n is the current number of iterations, N is the total number of iterations, F best is the fitness value of the population individual with the smallest fitness value in the current iteration, F agvn is the average fitness value of the population in the current iteration, and μ is the influencing factor; Among them, the adaptive calculation formula of mutation probability is the same as the crossover operation: Among them, P m is the probability of mutation in the next generation of the population, P mmax is the maximum mutation probability, P mx is greater than 0 and less than P mmax υ is the impact factor.

2. The method according to claim 1, characterized in that The neural network model is established based on the acquired flight body's angle of attack, flight Mach number, elevator angle and aerodynamic coefficient, including: Establishing a nonlinear identification model based on the angle of attack, the flight Mach number, the elevator deflection angle, and the aerodynamic coefficient; A neural network is used to fit the nonlinear identification model to establish the neural network model.

3. The method according to claim 2, characterized in that Using a neural network to fit the nonlinear identification model to establish the neural network model includes: Using the angle of attack, the flight Mach number and the elevator angle as three neurons of the input layer of the neural network; Using the aerodynamic coefficients as neurons of the output layer of the neural network, wherein the aerodynamic coefficients include the drag coefficient, lift coefficient, pitch moment coefficient, and relative position coefficient of the pressure center to the mass center of the flying body; Based on the setting of the neurons in the input layer and the output layer of the neural network, the neural network model is trained so that the neural network model fits the nonlinear identification model to establish the neural network model.

4. The method according to claim 1, wherein Optimizing the initial weights and thresholds of the neural network model using an adaptively adjusted genetic algorithm includes: Encoding the initial weights and thresholds of the neural network model to generate an initial population, and using the initial population as the current population to be optimized; The following steps are executed repeatedly until the number of iterations reaches the preset total number of iterations: Calculating the fitness value of the individuals in the population to be optimized, and determining whether the fitness value meets the optimization goal; If the fitness value meets the optimization goal, the initial weights and thresholds of the optimized neural network model are obtained, and the loop is exited. Otherwise, the population to be optimized is selected, cross-mutated, and a new generation of population is generated. The new generation of population is used as the population to be optimized, and the number of iterations is increased by 1.

5. The method according to claim 4, characterized in that Encoding the initial weights and thresholds of the neural network model to generate an initial population includes: selecting a real number encoding method to encode the initial weights and thresholds of the neural network model to form chromosome vectors of individuals in the initial population, and determining the initial population, wherein the initial population includes multiple individuals, and the chromosome vector of each individual includes all weights and thresholds of the neural network model.

6. The method according to claim 4, characterized in that Selecting the population to be optimized and performing crossover mutation processing to generate a new generation of population includes: Determining the probability of the population individual being selected according to the fitness value of the population individual to be optimized, and selecting a target individual from the population to be optimized; Adaptively adjusting the crossover probability, and performing a crossover operation on the chromosome vector of the target individual using the adaptively adjusted crossover probability; The mutation probability is adaptively adjusted, and the adaptively adjusted mutation probability is used to perform a mutation operation on the chromosome vectors of the individuals in the population to be optimized after the crossover operation, so as to generate a new generation of population.

7. An aerodynamic parameter identification device based on adaptive genetic algorithm and neural network, characterized in that: include: An establishing module is configured to establish a neural network model based on the acquired angle of attack, flight Mach number, elevator deflection angle, and aerodynamic coefficient of the flying body; an optimization module configured to optimize the initial weights and thresholds of the neural network model using an adaptively adjusted genetic algorithm, and train the neural network model based on the optimized initial weights and thresholds; an identification module configured to identify the real-time aerodynamic coefficients of the flying object using the optimized trained neural network model, and to identify the real-time aerodynamic derivatives of the flying object using the derivative characteristics of the neural network model; Wherein, the aerodynamic parameters include the aerodynamic coefficient and the aerodynamic derivative; The adaptive adjustment of the crossover probability includes: increasing the crossover probability of a population with a poor average fitness value in the next iteration during each iteration, reducing the crossover probability of a population with a good average fitness value in the next iteration, and gradually reducing the overall crossover probability as the number of iterations increases; and / or the adaptive adjustment of the mutation probability includes: increasing the mutation probability of a population with a poor average fitness value in the next iteration during each iteration, reducing the mutation probability of a population with a good average fitness value in the next iteration, and gradually reducing the overall mutation probability as the number of iterations increases; The calculation formula for the crossover probability is as follows: Among them, P c is the crossover probability of the next generation of the population, P c max is the maximum crossover probability set, P cx For a value greater than 0 and less than P c max The number of iterations, n is the current number of iterations, N is the total number of iterations, F best is the fitness value of the population individual with the smallest fitness value in the current iteration, F agvn is the average fitness value of the population in the current iteration, and μ is the influencing factor; Among them, the adaptive calculation formula of mutation probability is the same as the crossover operation: Among them, P m is the probability of mutation in the next generation of the population, P mmax is the maximum mutation probability, P mx is greater than 0 and less than P mmax υ is the impact factor.

8. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed, the computer is caused to execute the method according to any one of claims 1 to 6.

Images