A method for evaluating anti-ship missile attack time interval based on BP neural network

Abstract

The application discloses a kind of anti-ship missile attack time interval evaluation method based on BP neural network, comprising the following steps: S1, establish simulation model, obtain neural network training sample according to simulation model;S2, training sample is used to train BP neural network, and obtain guidance time prediction model;S3, desired range, maximum overload capacity, target field of view angle, target field of view angle maximum value are input into guidance time prediction model, and the guidance time range is obtained.The anti-ship missile attack time interval evaluation method based on BP neural network disclosed in the application does not need complicated calculation, can conveniently know the feasible guidance time of multiple missiles at this time, and the guidance time range obtained has high accuracy and short time consumption.

Description

TECHNICAL FIELD

[0001] The application relates to a BP neural network-based anti-ship missile attack time interval evaluation method and belongs to the field of guidance control. BACKGROUND

[0002] In the background of modern war, since many high-value battleships are equipped with strong penetration shipborne self-defense systems, single missile one-on-one attack faces great challenges. In order to break through the strong self-defense system, the guidance technology of cooperative attack has been rapidly developed. That is, many missiles hit a battleship at the same time, although their initial positions are different.

[0003] In most cases, each missile participating in cooperative attack is guided according to the expected attack time specified before launch. The expected guidance time is a necessary condition for realizing missile salvo attack. However, in the actual flight process, the missile is often subject to constraints such as the geometric relative relationship between the missile and the target, the missile overload, the target field of view angle, etc., so that the feasible range of the expected attack time is limited. Therefore, the accurate feasible range is an important link for realizing salvo attack. At present, formula is mostly used for solving, although the calculation is simple, but the precision is poor and the applicable range is limited.

[0004] Due to the above reasons, it is necessary to propose an anti-ship missile attack time interval evaluation method with high accuracy and convenient use. SUMMARY

[0005] In order to overcome the above problems, the present application has been designed, and a BP neural network-based anti-ship missile attack time interval evaluation method is designed, which comprises the following steps:

[0006] S1, a simulation model is established, and neural network training samples are obtained according to the simulation model;

[0007] S2, the BP neural network is trained by using the training samples, and a guidance time prediction model is obtained;

[0008] S3, the expected range, the maximum overload capacity, the target field of view angle and the maximum target field of view angle are input into the guidance time prediction model, and a guidance time range is obtained.

[0009] Further, in S1, the simulation model is obtained through the following sub-steps:

[0010] S11, a kinematic model and a time-controllable guidance law are established;

[0011] S12, the kinematic model and the time-controllable guidance law are solved by using the dichotomy method, and the simulation model is obtained.

[0012] Further, the input parameters of the simulation model are desired range, maximum overload capacity, target field of view angle, maximum target field of view angle, and the output is a guidance time range.

[0013] In a preferred embodiment, in S11, the kinematic model is represented as:

[0014]

[0015]

[0016]

[0017]

[0018]

[0019] wherein R is the missile-target distance, θ is the velocity pre-angle in the pitch direction, φ is the velocity pre-angle in the yaw direction, is the horizontal missile-target line-of-sight angle, is the vertical missile-target line-of-sight angle, and V is the missile speed.

[0020] a y , a z is a time-controllable guidance law instruction, wherein a y represents the lateral overload instruction, a z represents the longitudinal overload instruction.

[0021] In a preferred embodiment, in S11, the time-controllable guidance law a={a y , a z} is represented as:

[0022]

[0023] wherein e a is the unit vector of the lateral and longitudinal overload instructions,

[0024] σ is the target field of view angle, σ max is the maximum target field of view angle, K is the gain coefficient, N is the guidance coefficient, φ(x) is the shaping velocity angle function, t go is the remaining flight time, ε t is the attack time error.

[0025] In a preferred embodiment, in S12, in the bisection method, the overload buffeting position is taken as the boundary point of the guidance time.

[0026] In a preferred embodiment, S1 further has the step of:

[0027] S13. Sampling the input parameters at certain intervals to obtain input samples, inputting the input samples into the simulation model to obtain the corresponding guidance time boundary, and forming a training sample by the input samples and the corresponding guidance time boundary.

[0028] In a preferred embodiment, the BP neural network has three hidden layers, and each layer has five neurons.

[0029] In a preferred embodiment, in the BP neural network, the activation function adopts the sigmoid function.

[0030] In a preferred embodiment, when training the BP neural network, a mean square error function is used as a loss function.

[0031] The beneficial effects of the present invention include:

[0032] (1) According to the anti-ship missile attack time interval evaluation method based on BP neural network provided by the present invention, the feasible guidance time of multiple missiles at this time can be conveniently obtained without complicated calculations;

[0033] (2) The anti-ship missile attack time interval evaluation method based on BP neural network provided by the present invention uses a small number of calculation examples and is time-consuming;

[0034] (3) According to the anti-ship missile attack time interval evaluation method based on BP neural network provided by the present invention, the obtained guidance time range has high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A schematic flow chart of a method for evaluating an anti-ship missile attack time interval based on a BP neural network according to a preferred embodiment of the present invention is shown;

[0036] Figure 2 The ballistic trajectory simulation results in Example 1 are shown;

[0037] Figure 3 The simulation results of the attack time error in Example 1 are shown;

[0038] Figure 4 The overload instruction simulation results when the guidance time is at the inner edge of the upper and lower bounds in Example 2 are shown;

[0039] Figure 5 The simulation results of the trajectory when the guidance time is within the upper and lower bounds in Example 2 are shown;

[0040] Figure 6 The overload instruction simulation results when the guidance time is outside the upper and lower boundary edges in Example 2 are shown;

[0041] Figure 7Trajectory simulation results of the embodiment 2 when the guiding time is outside the upper and lower boundary edges. DETAILED DESCRIPTION

[0042] The application will be further described below by the accompanying drawings and embodiments. The features and advantages of the application will become more apparent through these descriptions.

[0043] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other implementations. Unless specifically stated otherwise, the drawings are not drawn to scale and the disclosure is not limited to the specific embodiments illustrated in the drawings.

[0044] According to the application, a BP neural network-based anti-ship missile attack time interval evaluation method is provided, as shown in the figure, comprising the following steps: Figure 1

[0045] S1, establishing a simulation model, obtaining neural network training samples according to the simulation model;

[0046] S2, training the BP neural network using the training samples to obtain a guiding time prediction model;

[0047] S3, inputting the expected range, maximum overload capacity, target field of view angle, and maximum target field of view angle into the guiding time prediction model to obtain a guiding time range.

[0048] The guiding time range can be represented by the upper limit and the lower limit of the feasible interval, i.e., the upper boundary point and the lower boundary point of the guiding time range.

[0049] Preferably, it further comprises S4, obtaining the guiding time ranges of multiple missiles, selecting a certain guiding time in the overlapping area as the common guiding time of the multiple missiles, thereby realizing the effect of multiple missiles attacking the target at the same time.

[0050] Further preferably, when the guiding time taken by the missile launching device exceeds the guiding time range, a warning is issued.

[0051] In the application, the BP neural network is applied, and the feasible range of the guiding time of multiple missiles can be quickly solved, thereby comprehensively selecting the guiding time in the same feasible range to realize the simultaneous attack of multiple missiles.

[0052] However, how to obtain the training samples of the BP neural network is the difficulty of the application.

[0053] According to the application, in S1, the simulation model is obtained through the following sub-steps:

[0054] ​S11, establishing a kinematic model and a time-controllable guidance law;

[0055] Further, a kinematic model and a time-controllable guidance law are established by the range, the maximum overload capacity, the target field of view angle and the maximum target field of view angle.

[0056] Further, in S11, the kinematic model is expressed as:

[0057]

[0058]

[0059]

[0060]

[0061]

[0062] wherein R is the missile-target distance, i.e. the range, θ is the velocity pre-angle in the pitch direction, φ is the velocity pre-angle in the yaw direction, is the horizontal missile-target line of sight angle, is the vertical missile-target line of sight angle, and V is the missile speed.

[0063] a y , a z is a time-controllable guidance law, wherein a y represents the lateral overload command, a z represents the longitudinal overload command.

[0064] Further, in S11, the time-controllable guidance law a = {a y , a z} is expressed as:

[0065]

[0066] wherein e a is the unit vector of the lateral and longitudinal overload command,

[0067] σ is the target field of view angle, σ max is the maximum target field of view angle, K is the gain coefficient, is a constant, N is the guidance coefficient, is a constant, φ(x) is the velocity angle shaping function, which can be set according to experience by those skilled in the art, t go is the remaining flight time, ε t is the attack time error.

[0068] wherein,

[0069] cosσ = cosθcosφ

[0070] e a = [sinφ / sinσ, sinθcosφ / sinσ] T

[0071] Further, for t go , ε t can be expressed as follows:

[0072]

[0073] ε t = t d -t-t go

[0074] wherein t d is the desired guidance time, and t is the current flight time.

[0075] Although the guidance time under the conditions that the different desired ranges, the maximum overload capacity, the target field of view angle and the maximum value thereof are known can be obtained according to the above kinematic model and the time-controllable guidance law, the kinematic model and the time-controllable guidance law can only determine whether the set guidance time is feasible, and the range of the guidance time cannot be obtained, which leads to that the effective neural network training samples cannot be formed only by the kinematic model and the time-controllable guidance law.

[0076] The inventor finds that when the selected guidance time exceeds the range, the overload will have a chattering phenomenon when the bisection method is used for solving; if within the range, the overload will not have the above-mentioned situation, based on the phenomenon, the inventor sets the overload chattering as a critical point, and the critical point is the boundary of the guidance time. That is, on both sides of the boundary point, the overload has different changes, and the problem that the kinematic model and the time-controllable guidance law cannot obtain the range of the guidance time is solved by the bisection method.

[0077] According to the present application, S12, the kinematic model and the time-controllable guidance law are solved by the bisection method to obtain a simulation model.

[0078] Further, the input parameters of the simulation model are the desired range, the maximum overload capacity, the target field of view angle and the maximum value of the target field of view angle, and the output is the range of the guidance time.

[0079] In a preferred embodiment, in S12, the overload chattering position is taken as the boundary point of the guidance time in the bisection method solving.

[0080] The inventor finds that when the selected guiding time exceeds the range, the overload will have a chattering phenomenon when solving by dichotomy; if within the range, the overload will not have the above-mentioned situation, based on the phenomenon, the inventor sets the overload chattering as a critical point, which is the boundary of the guiding time. That is, on both sides of the boundary point, the overload has different changes, and the dichotomy solves the problem that the kinematic model and the time-controllable guiding law cannot obtain the guiding time range.

[0081] According to the application, the batch random setting input parameters are input into the simulation model, so that the batch simulation data can be obtained.

[0082] The inventor finds that although the simulation model can obtain the batch simulation data, directly taking the batch simulation data as the sample of the neural network will cause problems such as too many samples and large calculation amount.

[0083] In a preferred embodiment, the optimal input sample is also obtained through sensitivity analysis, so as to reduce the calculation amount under the premise of ensuring the accuracy. In the application, the sensitivity analysis refers to comparing and analyzing the batch simulation data according to the simulation model to find the data with typical representation as the sample of the neural network.

[0084] Specifically, S1 also has the following steps:

[0085] S13, within the selectable range of the input parameters, the input parameters are sampled at certain intervals to obtain input samples, the input samples are input into the simulation model to obtain the boundary of the corresponding guiding time, and the training sample is composed of the input samples and the corresponding guiding time boundary.

[0086] Further, since the input parameters have multiple, only one parameter varies according to the interval and the remaining parameters are fixed during each sampling process, so as to obtain more comprehensive input samples.

[0087] The inventor finds through a large amount of research and analysis that sampling the range parameter at an interval of 200 m, sampling the maximum overload capacity at an interval of 1G, sampling the target field of view angle at an interval of 2deg, and sampling the maximum value of the target field of view angle at an interval of 2deg can obtain the most comprehensive training sample, so as to train the optimal guiding time prediction model.

[0088] According to a preferred embodiment of the application, the hidden layer of the BP neural network is 3 layers, and the number of neurons in each layer is 5.

[0089] The inventor determines through a large amount of experiments that the above-mentioned parameters can greatly improve the prediction accuracy and the prediction speed is relatively fast.

[0090] Further preferably, in the BP neural network, the activation function adopts a sigmoid function, which is easy to derive and facilitates improving the operation efficiency.

[0091] In a preferred embodiment, 70% of the samples are selected from the training samples as a training set, 15% of the samples are selected as a test set, and 15% of the samples are selected as a validation set.

[0092] The learning process of the BP neural network consists of two processes, i.e., signal forward propagation and error back propagation. In the forward propagation, the input sample is transmitted from the input layer to the output layer through the processing of each hidden layer. If the actual output of the output layer does not match the expected output, the error back propagation stage is entered. In the back propagation, the output is transmitted to the input layer through the hidden layer in a certain form, and the error is allocated to all units of each layer, so as to obtain the error signal of each unit, which is used as the basis for correcting the weight of each unit.

[0093] For the forward propagation, the final output value and the loss value between the output value and the actual value are calculated according to the input sample, the given initial weight value w and the bias value b in this process. Preferably, the calculation formula between the input and the output is as follows:

[0094]

[0095] Wherein, f() is an activation function, y is an output, x is an input, and n is a number of layers.

[0096] According to a preferred embodiment of the present application, when the BP neural network is trained, a mean square error function is used as a loss function.

[0097] The mean square error function is expressed as:

[0098]

[0099] Wherein, n is a number of layers, y is an output value, is an expected output value.

[0100] Embodiment

[0101] Embodiment 1

[0102] The guide time prediction model is obtained by the following steps:

[0103] S1, a simulation model is established, and neural network training samples are obtained according to the simulation model;

[0104] S2, the BP neural network is trained by using the training samples to obtain a guide time prediction model;

[0105] In S1, the simulation model is obtained by the following sub-steps:

[0106] S11, establishing a kinematic model and a time-controllable guidance law;

[0107] S12, solving the kinematic model and the time-controllable guidance law by using a dichotomy method to obtain a simulation model;

[0108] S13, sampling input parameters at a predetermined interval to obtain input samples, inputting the input samples into the simulation model to obtain a boundary of a corresponding guidance time, and grouping the input samples and the corresponding guidance time boundary to form a training sample, specifically, sampling a range parameter at an interval of 200 m, sampling a maximum overload capacity at an interval of 1 G, sampling a target field of view angle at an interval of 2 deg, and sampling a maximum value of the target field of view angle at an interval of 2 deg.

[0109] The input parameters of the simulation model are a desired range, a maximum overload capacity, a target field of view angle, and a maximum value of the target field of view angle, and the output is a guidance time range

[0110] In S11, the kinematic model is represented as:

[0111]

[0112]

[0113]

[0114]

[0115]

[0116] The time-controllable guidance law a = {a y , a z} is represented as:

[0117]

[0118] wherein cosσ = cosθcosφ

[0119] e a = [sinφ / sinσ, sinθcosφ / sinσ] T

[0120]

[0121] ε t = t d -t-t go

[0122] In S2, the hidden layer of the BP neural network is 3 layers, the number of neurons in each layer is 5, the activation function adopts a sigmoid function, and when the BP neural network is trained, a mean square error function is used as a loss function.

[0123] S3, input the expected range, maximum overload capacity, target field of view angle, target field of view angle maximum value into the guidance time prediction model to obtain a guidance time range; in the embodiment, the step is performed through a simulation experiment, and three missiles are simulated to be launched at the same time, wherein the speeds are all 250 m / s, N in the overload instruction is all 3, K is all 8, the missile launch positions are (5800, 0, 0), (7000, 0, 0) and (6200, 0, 0) respectively, the target position is (0, 0, 0), and the missile expected range, maximum overload capacity, target field of view angle and target field of view angle maximum value parameters are as shown in Table 1, the parameters are input into the guidance time prediction model, and the obtained guidance time range is as shown in Table 2.

[0124] Table 1

[0125]

[0126] Table 2

[0127]

[0128] According to the obtained guidance time range, it can be seen that the three missiles have a common flyable range. Therefore, the three missiles can realize salvo attack, the guidance time is selected as 42 s, simulation is performed, the trajectory of the trajectory is as shown in Figure 2 , Figure 3 and the attack time error is as shown in .

[0129] From Figure 2 , 3 it can be seen that under different initial conditions, the three missiles can hit the target at the same time and complete salvo attack.

[0130] Embodiment 2

[0131] The guidance time prediction model is obtained through the following steps:

[0132] S1, a simulation model is established, and neural network training samples are obtained according to the simulation model;

[0133] S2, the BP neural network is trained by using the training samples to obtain the guidance time prediction model;

[0134] In S1, the simulation model is obtained through the following sub-steps:

[0135] S11, a kinematics model and a time-controllable guidance law are established;

[0136] S12, the kinematic model and the time-controllable guidance law are solved by dichotomy to obtain a simulation model;

[0137] S13, input parameters are sampled at a predetermined interval to obtain input samples, the input samples are input into the simulation model to obtain corresponding guidance time boundaries, the training samples are composed of the input samples and the corresponding guidance time boundaries, specifically, the range parameter is sampled at an interval of 200 m, the maximum overload capacity is sampled at an interval of 1 G, the target field of view angle is sampled at an interval of 2 deg, and the maximum target field of view angle is sampled at an interval of 2 deg.

[0138] The input parameters of the simulation model are expected range, maximum overload capacity, target field of view angle, and maximum target field of view angle, and the output is a guidance time range

[0139] In S11, the kinematic model is represented as:

[0140]

[0141]

[0142]

[0143]

[0144]

[0145] The time-controllable guidance law a = {a y , a z} is represented as:

[0146]

[0147] Wherein, cosσ = cosθcosφ

[0148] e a = [sinφ / sinσ, sinθcosφ / sinσ] T

[0149]

[0150] ε t = t d -t-t go

[0151] In S2, the hidden layer of the BP neural network is 3 layers, the number of neurons in each layer is 5, and the activation function adopts a sigmoid function, and when the BP neural network is trained, a mean square error function is used as a loss function.

[0152] S3, input the expected range, maximum overload capacity, target field of view angle, target field of view angle maximum value into the guidance time prediction model to obtain a guidance time range; in this embodiment, this step is performed through simulation experiment, a missile is simulated to be launched, the missile speed is 250 m / s, in the overload instruction, N is 3, K is 8, the missile launch position is (8000, 0), the target point coordinates are (0, 0, 0), the missile expected range, maximum overload capacity, target field of view angle, target field of view angle maximum value parameters are shown in Table Three, the parameters are input into the guidance time prediction model, and the obtained guidance time range is shown in Table Four.

[0153] Table Three

[0154]

[0155] Table Four

[0156]

[0157] The guidance time of the upper and lower boundary edges is selected, that is, 46 s and 64 s, simulation is performed, the overload instruction in the simulation result is shown in Figure 4 , and the trajectory is shown in Figure 5 .

[0158] The guidance time close to the outside of the upper and lower boundary edges is selected, that is, 45 s and 65 s, simulation is performed, the overload instruction in the simulation result is shown in Figure 6 , and the trajectory is shown in Figure 7 .

[0159] As can be seen from Figure 4 , Figure 5 , when the guidance time is located in the upper and lower boundary edges, the missile can successfully hit the target, and the overload instruction does not have chattering phenomenon. As can be seen from Figure 6 , Figure 7 , when the guidance time is located outside the upper and lower boundary edges, the overload instruction has chattering phenomenon, especially when 45 s, the chattering phenomenon is serious, according to actual experience, when the overload instruction has chattering phenomenon, the missile cannot accurately hit the target.

[0160] As can be seen from the above, when the selected guidance time is in the obtained guidance time range, the target can be successfully hit and the overload instruction does not have chattering; if not in the obtained guidance time range, the overload instruction has chattering, which leads to the target cannot be hit, which shows that the upper and lower boundaries of the obtained guidance time range in this embodiment are accurate.

[0161] In the description of the present application, it should be noted that the terms "upper", "lower", "inner", "outer", "front", "back" and the like indicate the positional or location relationship based on the working state of the present application, and are only for the purpose of facilitating the description of the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", "third", "fourth" are only for the purpose of description, and cannot be understood as indicating or implying relative importance.

[0162] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium; it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0163] The above describes the present application in combination with the preferred embodiments, but these embodiments are only exemplary and serve only to illustrate. On this basis, various substitutions and improvements can be made to the present application, which all fall within the scope of protection of the present application.

Claims

1. A method for evaluating the attack time interval of an anti-ship missile based on a BP neural network, characterized in that: The following steps are involved: S1. Establish a simulation model and obtain neural network training samples based on the simulation model; S2. Using training samples to train the BP neural network to obtain a guidance time prediction model; S3. Input the expected range, maximum overload capacity, target field of view angle, and maximum target field of view angle into the guidance time prediction model to obtain the guidance time range; In S1, the simulation model is obtained through the following sub-steps: S11. Establish kinematic model and time-controllable guidance law; S12, using a bisection method to solve the kinematic model and the time-controllable guidance law to obtain a simulation model; In S11, the kinematic model is expressed as: Where R is the missile-target distance, θ is the velocity lead angle in the pitch direction, and φ is the velocity lead angle in the yaw direction. is the horizontal sight angle of the missile, is the vertical sight angle between the missile and the target, and V is the missile speed; a y 、a z is a time-controllable guidance law instruction, where a y Characterizes the horizontal overload instruction, a z Characterizes the longitudinal overload instruction; In S11, the time controllable law a={a y , a z } is represented as: Among them, e a is the unit vector of the horizontal and vertical overload instructions, σ is the target field of view angle, σ max is the maximum value of the target field of view angle, K is the gain coefficient, N is the guidance coefficient, φ(x) is the shaping velocity angle function, t go is the remaining flight time, ε t is the attack time error.

2. The anti-ship missile attack time interval evaluation method based on BP neural network according to claim 1 is characterized in that: The input parameters of the simulation model are expected range, maximum overload capacity, target field of view angle, and maximum target field of view angle, and the output is the guidance time range.

3. The anti-ship missile attack time interval evaluation method based on BP neural network according to claim 1 is characterized in that: In S12, in the binary solution, the overload buffeting position is used as the boundary point of the guidance time.

4. The anti-ship missile attack time interval evaluation method based on BP neural network according to claim 1 is characterized in that: S1 also has the steps: S13. Sampling the input parameters at certain intervals to obtain input samples, inputting the input samples into the simulation model to obtain the corresponding guidance time boundary, and forming a training sample by the input samples and the corresponding guidance time boundary.

5. The anti-ship missile attack time interval evaluation method based on BP neural network according to claim 1 is characterized in that: The BP neural network has three hidden layers, and each layer has five neurons.

6. The anti-ship missile attack time interval evaluation method based on BP neural network according to claim 1 is characterized in that: In BP neural network, the activation function uses sigmoid function.

7. The anti-ship missile attack time interval evaluation method based on BP neural network according to claim 1 is characterized in that: When training the BP neural network, the mean square error function is used as the loss function.

Images