Snap bomb opening point speed error prediction method based on SSA-BP neural network

Abstract

The invention provides a shrapnel opening point speed error prediction method based on an SSA-BP neural network, relates to the technical field of projectile outer trajectory parameter calculation, and establishes an opening point speed error prediction model of an SSA algorithm optimized BP neural network for solving the problem that a BP prediction algorithm is caught in local optimum due to improper initial weight and threshold values. Taking the initial velocity, the launching angle, the X-direction wind speed condition and the Z-direction wind speed condition of the shrapnel as input indexes, and taking the X-direction speed error and the Z-direction speed error of the opening point of the shrapnel as output indexes; and acquiring a speed error data set of the opening point of the shrapnel, carrying out data preprocessing, and loading the data set into the SSA-BP prediction model, so that the X-direction speed error and the Z-direction speed error of the opening point of the shrapnel can be directly predicted.

Description

Technical Field

[0001] The invention relates to the technical field of projectile outer ballistic parameter calculation, and in particular to a method for predicting the speed error of a submunition opening point based on an SSA-BP neural network. Background Art

[0002] In modern military conflicts, precise firepower strikes on long-range military targets have become a key factor affecting the outcome of military conflicts. Many countries regard long-range precision firepower strikes as an important development direction of ammunition technology. To ensure that the projectile can hit the target at a long distance, it is necessary to predict the projectile's related external ballistic parameters, adjust the projectile launch mechanism according to the difference between the predicted result and the expected result, correct the projectile's external ballistics, and thus achieve precise firepower strikes.

[0003] The speed of the opening point of the cluster bomb in the X and Z directions affects the dispersion and strike performance of the sub-killing element. Therefore, the accurate prediction of the speed error of the opening point of the cluster bomb is crucial for the cluster bomb to achieve long-range precision firepower strike. However, in the actual process, the processing power of the fire control computer is limited, the solution process is allowed to take a short time, and the sensor output data has a certain deviation from the true value, so the ballistic solution error is inevitable. However, it is very effective to use a high-precision prediction model to predict the speed error of the opening point of the cluster bomb in the X and Z directions.

[0004] The current patent disclosure shows that the patent (CN 112379364 A) discloses a method for high-precision prediction of projectile impact velocity based on a neural network radar combination. This method uses the impact velocity of the projectile tested by the continuous wave radar as the output signal, and the radial velocity of the projectile before the impact moment tested by the initial velocity radar as the input signal to establish a BP neural network prediction model; the BP neural network model is established using the impact velocity data of the projectile that has been tested by the two radars, and the impact velocity of the projectile that has not been measured by the continuous wave radar can be predicted. The accuracy of this method reaches within 2‰, which is far higher than the accuracy requirement of the impact velocity itself, but the process of this method is relatively cumbersome, time-consuming, requires a large amount of test data, and is costly.

[0005] In recent years, more and more swarm intelligence optimization algorithms have been applied to the fields of system control, pattern recognition, performance prediction, etc. Swarm intelligence optimization algorithm is a random search algorithm based on probability theory, which mainly simulates the unique behavior and survival mode of biological groups in the process of evolution, such as the foraging behavior of insects, birds and other groups; the group cooperates in a certain way to find food, and can obtain more food faster through communication. The algorithm abstracted from this group behavior is called swarm intelligence optimization algorithm. The present invention establishes a prediction model of the speed error of the opening point of the cluster bomb by optimizing the BP neural network with the sparrow search algorithm, which is used to predict the relationship between the projectile launch parameters and meteorological conditions and the speed error of the opening point of the cluster bomb, and compares and analyzes the prediction model of the BP neural network optimized by the traditional BP, PSO algorithm and GA algorithm, and verifies the prediction effect of the SSA-BP model. This method can directly predict the speed error of the opening point of the cluster bomb in the X and Z directions, improves the efficiency of obtaining the speed error of the opening point of the cluster bomb in the X and Z directions, and provides convenience for the research of long-range firepower planning. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a method for predicting the speed error of the opening point of a cluster bomb based on an SSA-BP neural network, which solves the problems of poor computing power and low prediction accuracy of the traditional method.

[0007] A method for predicting the speed error of a cluster bomb opening point based on an SSA-BP neural network comprises the following steps:

[0008] Step S1: Establish an XZ rectangular coordinate system with the muzzle as the coordinate origin, the X direction is the projectile launch direction, the angle between the flight direction and the X direction is the launch angle, and the Z direction is the vertical direction of the horizontal plane; select different submunition initial velocities, launch angles, X-direction wind speed conditions and Z-direction wind speed conditions, obtain the speed error data sets of the submunition opening point in the X and Z directions by calculation, and normalize the data in the speed error data set;

[0009] The normalization process uses the mapminmax function to map the data to the interval [-1,1]. The specific formula is as follows:

[0010]

[0011] Where: x max is the maximum value of the sample data; x min is the minimum value of the sample data; y max =1;y min = -1; x and x * They are the data before and after normalization respectively;

[0012] Step S2: taking the initial velocity of the cluster bomb, the launch angle, the wind speed condition in the X direction and the wind speed condition in the Z direction as input indicators, and taking the speed error in the X direction and the speed error in the Z direction of the cluster bomb opening point as output indicators, using the SSA algorithm to optimize the BP neural network, and establishing an SSA-BP neural network model, that is, a speed error prediction model for the opening point;

[0013] The SSA-BP neural network model adopts a three-layer structure, including an input layer, a hidden layer and an output layer, wherein the input layer is four parameters, namely, the initial velocity of the cluster bomb, the launch angle, the wind speed condition in the X direction and the wind speed condition in the Z direction, and the output layer is two parameters, namely, the speed error in the X direction and the speed error in the Z direction of the cluster bomb opening point; through search and trial calculation, it is determined that the number of neuron nodes in the hidden layer is 13, the activation function of the hidden layer adopts the logsig function, and the activation function of the output layer adopts the pureline function;

[0014] Step S3: The uncalculated initial velocity of the cluster bomb, the launch angle, the wind speed condition in the X direction, and the wind speed in the Z direction are brought into the SSA-BP neural network model as input indicators to predict the corresponding speed error in the X direction and the speed error in the Z direction of the cluster bomb opening point.

[0015] The beneficial effects of adopting the above technical solution are:

[0016] The present invention provides a method for predicting the speed error of a submunition opening point based on an SSA-BP neural network. Based on the consideration of the constant wind speed exterior ballistics, the present invention establishes an SSA-BP submunition opening point speed error prediction model with the input layer being the initial velocity, the firing angle, the wind speed in the X direction and the wind speed in the Z direction, and the output layer being the speed error in the X direction and the speed error in the Z direction of the opening point, and directly predicts the speed error of the opening point; the positive effect of the present invention is to fill the gap in the prediction method in the current field of submunition exterior ballistics, effectively solve the problem of the difficulty in evaluating the speed error of the submunition opening point, and a very accurate prediction value can be provided by this method; the four model prediction values ​​are compared and analyzed with the true values, and the results prove that the prediction accuracy of the SSA-BP neural network model is the highest, and the prediction accuracy fully meets the requirements, and the present invention can provide assistance for the planning of firepower strike programs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a structural diagram of the SSA-BP neural network model in an embodiment of the present invention;

[0018] Figure 2 Flowchart of BP, PSO-BP, GA-BP and SSA-BP prediction models. DETAILED DESCRIPTION

[0019] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0020] A method for predicting the speed error of a cluster bomb opening point based on an SSA-BP neural network comprises the following steps:

[0021] Step S1: Establish an XZ rectangular coordinate system with the muzzle as the coordinate origin, the X direction is the projectile launch direction, the angle between the flight direction and the X direction is the launch angle, and the Z direction is the vertical direction of the horizontal plane; select different submunition initial velocities, launch angles, X-direction wind speed conditions and Z-direction wind speed conditions, obtain the speed error data sets of the submunition opening point in the X and Z directions by calculation, and normalize the data in the speed error data set;

[0022] Ballistic dispersion will affect the range of the projectile, and ballistic dispersion is caused by factors such as the firing angle, initial velocity and meteorological wind conditions during the launch process. Therefore, the projectile's charge structure, projectile-gun interaction, chamber pressure characteristics, random vibration of the launch mechanism during shooting and wind force level will affect the projectile range. The present invention only considers the influence of the external ballistic initial velocity, firing angle and wind conditions on the projectile range, and uses MATLAB software to perform external ballistic calculations on the projectile. In this embodiment, the initial velocity of the submunition is taken every 5 m / s from 755 to 800 m / s, the firing angle is taken every 5° from 15° to 60°, the wind speed in the X direction is taken every 2 m / s from 2 to 10 m / s, and the wind speed in the Z direction is taken every 2 m / s from 2 to 10 m / s. MATLAB programming is used to adopt the Runge-Kutta method to solve the ballistic equation calculation. After multiple calculations, a total of 2500 groups of data are obtained, of which 2000 groups of data are used as network data training samples, and the remaining 500 groups of data are randomly selected as test samples;

[0023] The normalization process uses the mapminmax function to map the data to the interval [-1,1]. The specific formula is as follows:

[0024]

[0025] Where: x max is the maximum value of the sample data; x min is the minimum value of the sample data; y max =1;y min = -1; x and x * They are the data before and after normalization respectively;

[0026] Step S2: Taking the initial velocity of the cluster bomb, the launch angle, the wind speed condition in the X direction and the wind speed condition in the Z direction as input indicators, and the speed error in the X direction and the speed error in the Z direction of the cluster bomb opening point as output indicators, the SSA algorithm is used to optimize the BP neural network, and an SSA-BP neural network model is established, that is, a speed error prediction model for the opening point, such as Figure 1 As shown;

[0027] The SSA algorithm is an intelligent algorithm inspired by the sparrow's hunting and avoiding predators. SSA uses the foraging behavior and anti-predation behavior of the sparrow group as a bionic object, and has the characteristics of high accuracy, fast convergence speed, strong stability, and good global search ability. The algorithm implementation steps are as follows:

[0028] (1) Initialize the position and fitness of the sparrow population.

[0029] (2) By sorting the population, the best individual position and the best fitness value of the sparrow are obtained.

[0030] (3) Foraging behavior, updating the finder's location:

[0031]

[0032] Where: Q is a random number that obeys the normal distribution; L is a unit row vector; a is a random number between [0,1]; is the position information of the i-th sparrow in the j-th dimension after the t-th iteration; N and t are the maximum number of iterations and the current number of iterations respectively; R2 and S T They are warning value and safety value respectively.

[0033] (4) Update the joiner's location:

[0034]

[0035] Where: X P is the current discoverer's position; X worst is the global worst position; A + is a row vector containing only two random elements, 1 and -1.

[0036] (5) Anti-predation behavior, updating the location of the sparrow population:

[0037]

[0038] Where: K is a random number between [-1,1], f i is the individual fitness value, ε is a constant close to 0; X best is the current global optimal position; β is the step size parameter; f best and f worst are the global optimal fitness value and the global worst fitness value respectively.

[0039] (6) Update the historical optimal fitness;

[0040] (7) Repeat the above steps (2) to (6). When the maximum number of iterations is reached, the loop ends and the fitness value and the optimal individual position are output.

[0041] Although the BP neural network has strong nonlinear mapping capabilities, it is based on the gradient descent method to adjust the connection weight threshold of the network, thereby performing error back propagation, which can easily lead to falling into the local optimum and reduce the accuracy of prediction. The error value of the BP neural network is used as the fitness function of the SSA algorithm. The SSA algorithm is used to search and optimize to obtain the best weight and threshold, and then assign them to the BP neural network. The steps of the SSA-BP algorithm are as follows: Figure 1 The SSA-BP neural network model adopts a three-layer structure. The input layer is composed of four parameters: the initial velocity of the submunition, the launch angle, the wind speed condition in the X direction, and the wind speed condition in the Z direction. The output layer is composed of two parameters: the speed error in the X direction and the speed error in the Z direction of the submunition opening point. Through search and trial calculation, the number of hidden layer neuron nodes is determined to be 13. The activation function of the hidden layer adopts the logsig function, and the activation function of the output layer adopts the pureline function.

[0042] In the search trial, the number of hidden layer nodes is set in the search range of 2 to 98, and is substituted into the neural network in sequence with a step size of 1 for inspection and verification. The number of hidden layer nodes corresponding to the best fitting effect is selected as the number of hidden layer nodes of the BP neural network.

[0043] Step S3: The uncalculated initial velocity of the cluster bomb, the launch angle, the wind speed condition in the X direction, and the wind speed in the Z direction are brought into the SSA-BP neural network model as input indicators to predict the corresponding speed error in the X direction and the speed error in the Z direction of the cluster bomb opening point.

[0044] In this embodiment, BP, PSO-BP and GA-BP models are introduced as comparison models. The prediction accuracy of the three models is tested by using the above training data and the prediction accuracy of the three models is compared with the prediction accuracy of the SSA-BP neural network model to determine which model has the highest prediction accuracy. Figure 2 As shown in the figure, the setting of training parameters directly affects the performance of BP neural network, which is crucial. After repeated calculations, the settings are as follows:

[0045] The learning rate of the BP neural network is set to 0.01, the number of training times is set to 1000, and the expected training error is set to 0.00001. Since too many hidden layers may cause the error function to fail to converge or the convergence time is too long, and the BP network with a single hidden layer can approximate any continuous function in a closed interval, the BP neural network adopts a three-layer structure. The input layer is the four parameters of the initial velocity of the projectile, the launch angle, the wind speed condition in the X direction and the wind speed condition in the Z direction, and the output layer is the two parameters of the speed error in the X direction and the speed error in the Z direction of the submunition opening point. The number of hidden layer nodes is set to the search range of 2 to 98, and the neural network is substituted into the neural network in turn with a step size of 1 for inspection and verification. The number of hidden layer nodes corresponding to the best fitting effect is selected as the number of hidden layer nodes of the BP neural network in this paper. Finally, the number of hidden layer neuron nodes is determined to be 13 through search and trial calculation, the hidden layer activation function uses the logsig function, and the output layer activation function uses the pureline function. The network topology of the SSA-BP, PSO-BP, and GA-BP models is set to 4-13-2.

[0046] The initial population size of GA is set to 40, the maximum evolutionary generation is 60, the crossover probability is 0.7, the mutation probability is 0.6, and the chromosome code length is set to 3×13+13×1+13+1=66; the initial population size of PSO is set to 40, the maximum evolutionary generation is 60, the individual acceleration constant C1 is 1.5, the social acceleration constant C2 is 1.5, and the inertia weight is 0.8; the initial sparrow population of SSA is set to 40, the maximum number of iterations is 60, the flock warning value is 0.8, the proportion of discoverers is 0.4, and the proportion of sparrows aware of the danger is 0.3.

[0047] The data set was simulated and tested using MATLAB software combined with four prediction models. The average prediction errors of the BP model in the X and Z directions were 7.75% and 8.23%, the average prediction errors of the GA-BP model in the X and Z directions were 5.68% and 4.36%, the average prediction errors of the PSO-BP model in the X and Z directions were 4.23% and 5.06%, and the average prediction errors of the SSA-BP model in the X and Z directions were 2.79% and 2.51%, all below 3%, with the highest prediction accuracy.

[0048] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) to form a technical solution.

Claims

1. A method for predicting the speed error of the opening point of a cluster bomb based on an SSA-BP neural network, characterized in that: The following steps are involved: Step S1: Establish an XZ rectangular coordinate system with the muzzle as the coordinate origin, the X direction is the projectile launch direction, the angle between the flight direction and the X direction is the launch angle, and the Z direction is the vertical direction of the horizontal plane; select different submunition initial velocities, launch angles, X-direction wind speed conditions and Z-direction wind speed conditions, obtain the speed error data sets of the submunition opening point in the X and Z directions by calculation, and normalize the data in the speed error data set; Step S2: taking the initial velocity of the cluster bomb, the launch angle, the wind speed condition in the X direction and the wind speed condition in the Z direction as input indicators, and taking the speed error in the X direction and the speed error in the Z direction of the cluster bomb opening point as output indicators, using the SSA algorithm to optimize the BP neural network, and establishing an SSA-BP neural network model, that is, a speed error prediction model for the opening point; Step S3: The uncalculated initial velocity of the cluster bomb, the launch angle, the wind speed condition in the X direction, and the wind speed in the Z direction are brought into the SSA-BP neural network model as input indicators to predict the corresponding speed error in the X direction and the speed error in the Z direction of the cluster bomb opening point.

2. According to claim 1, a method for predicting the speed error of the opening point of a cluster bomb based on an SSA-BP neural network is characterized in that: The normalization process in step S1 uses the mapminmax function to map the data to the interval [-1,1]. The specific formula is as follows: Where: x max is the maximum value of the sample data; x min is the minimum value of the sample data; y max =1;y min = -1; x and x * They are the data before and after normalization respectively.

3. The method for predicting the speed error of the opening point of a cluster bomb based on an SSA-BP neural network according to claim 1, characterized in that: The SSA-BP neural network model in step S2 adopts a three-layer structure, including an input layer, a hidden layer and an output layer, wherein the input layer is four parameters of the initial velocity of the cluster bomb, the launch angle, the wind speed condition in the X direction and the wind speed condition in the Z direction, and the output layer is two parameters of the speed error in the X direction and the speed error in the Z direction of the opening point of the cluster bomb; Through search and trial calculation, it is determined that the number of neuron nodes in the hidden layer is 13, the hidden layer activation function uses the logsig function, and the output layer activation function uses the pureline function.

Images