A control method for accurate and rapid acquisition of parameters based on multi-network model

Abstract

The invention discloses a control method for accurately and quickly obtaining various elements based on a multi-network model, comprising the following steps: setting an element solving neural network model to obtain shooting elements in a windless state, wherein the shooting elements include a sight, a range and a ballistic height; setting a longitudinal wind neural network model to obtain the average longitudinal wind during the flight of a shell according to the sight and the ballistic height in the shooting elements in the windless state, as well as the actual wind speed and wind direction, as the average ballistic longitudinal wind; setting an element correction neural network model, taking the average ballistic longitudinal wind as input, outputting a sight correction amount, correcting the shooting elements in the windless state, and obtaining the final shooting elements; adjusting the launch platform according to the final shooting elements, and launching the shell. The control method for accurately and quickly obtaining various elements based on a multi-network model provided by the invention fully considers real-time meteorology, can improve the accuracy of element solving, has a small amount of calculation, is short in time consumption, does not require cumbersome firing table query, can be deployed in a fire control computer, and can effectively avoid missing the target.

Description

Technical Field

[0001] The invention relates to a control method for accurately and quickly obtaining various parameters based on a multi-network model, and belongs to the technical field of aircraft guidance control. Background Art

[0002] Semi-active laser terminally guided artillery shells are widely used because they combine the low cost of conventional artillery with the precision strike capability of missiles.

[0003] In the semi-active laser terminal guidance projectile, in order to capture the target during the terminal guidance, the firing table is currently used to determine the firing parameters, including the sight (firing angle), program (inertial gyro unlocking time), delay (laser irradiator start irradiation time) as the basic parameters, and the correction parameters are used to correct the meteorological deviation (including air pressure, temperature, wind direction, wind speed), etc. Whether the calculation of the parameters is reasonable is related to whether the subsystem can work normally, and directly affects the hit probability of this type of ammunition.

[0004] At present, most of the methods use the method of querying the shooting table to solve the parameters, but this method has the following problems: 1) This method uses linear interpolation to solve the shooting parameters, but the essence of this problem is a multi-factor coupling problem, so the accuracy is not high; 2) This method needs to interpret the shooting table to solve it, which leads to slow solution speed. Although some shooting parameter calculation software based on traditional shooting tables has been developed, many factors related to the parameters are ignored, and the solution accuracy is poor; 3) For meteorological information, if the layered meteorological information obtained in real time is considered, the scale of the shooting table will increase significantly, and the amount of interpolation calculation is large, which is difficult to meet the needs of fast calculation.

[0005] Due to the above reasons, the inventors have conducted extensive research on the existing semi-active laser terminal-guided artillery projectile guidance methods in order to propose a guidance control method that can solve the above problems. Summary of the invention

[0006] In order to overcome the above problems, the inventors have conducted in-depth research and designed a control method for accurately and quickly obtaining various parameters based on a multi-network model, comprising the following steps:

[0007] S1. Setting a neural network model for calculating various elements to obtain shooting elements in a windless state, wherein the shooting elements include sight, loading and trajectory height;

[0008] S2, setting up a longitudinal wind neural network model, and obtaining the average longitudinal wind during the flight of the artillery shell according to the sight and trajectory height in the windless state shooting parameters, as well as the actual wind speed and wind direction, as the average trajectory longitudinal wind;

[0009] S3, setting up a neural network model for correction of various elements, taking the average longitudinal wind of the trajectory as input, outputting the correction value of the sight, correcting various elements of shooting in a windless state, and obtaining the final shooting elements;

[0010] S4. Adjust the launch platform according to the final shooting parameters and launch the shells.

[0011] In a preferred embodiment, in S1, the air pressure and temperature in the actual weather are fitted, and the fitting parameters and the expected range are used as inputs of the neural network model for solving various parameters.

[0012] In a preferred embodiment, the air pressure fitting is performed using an exponential curve.

[0013] The temperature fitting adopts cubic function fitting.

[0014] In a preferred embodiment, the year is divided into multiple parts, and the air pressure and temperature in each time period are fitted separately to obtain multiple fitting parameters. For each fitting parameter, the neural network model for solving various parameters is trained independently.

[0015] In a preferred embodiment, the year is divided into four parts according to quarters. In actual use, the neural network model with various parameters trained with fitting parameters corresponding to the quarter of the current time is used for solution.

[0016] In a preferred embodiment, the neural network model for solving various variables adopts a BP neural network, uses two fully connected layers as hidden layers, and each layer has 10 neurons.

[0017] In a preferred embodiment, samples for training the neural network model for solving various parameters are obtained through ballistic trajectory simulation.

[0018] In a preferred embodiment, in S2, the longitudinal wind force magnitudes received by the shells at different altitudes, i.e., the longitudinal wind array, are obtained according to the wind speed, wind direction, gun position reference, and shell-target distance in the actual weather.

[0019] Set up a longitudinal wind neural network model, take the sight and trajectory height in the windless state shooting parameters as input, and output the longitudinal wind weight matrix at different altitudes;

[0020] The average ballistic longitudinal wind is obtained based on the longitudinal wind array and the longitudinal wind weight matrix.

[0021] In a preferred embodiment, 200 meters is taken as one altitude layer to obtain longitudinal wind arrays and longitudinal wind weight matrices at different altitudes.

[0022] In a preferred embodiment, in S3, the sight output by the neural network model for element solution is corrected with the sight correction amount to obtain a corrected sight, and the corrected sight and the loading and trajectory height output by the neural network model for element solution are used as the final shooting elements.

[0023] The beneficial effects of the present invention include:

[0024] (1) The control method for accurately and quickly obtaining various parameters based on the multi-network model provided by the present invention fully considers the real-time weather and can improve the accuracy of various parameter calculations;

[0025] (2) The control method for accurately and quickly obtaining various parameters based on the multi-network model provided by the present invention can be deployed in a fire control computer, has a small amount of calculation, is short in time, can be solved quickly, and does not require cumbersome firing table query;

[0026] (3) The control method for accurately and quickly obtaining the parameters based on the multi-network model provided by the present invention makes the landing point of the fired shell closer to the target when entering the laser guidance stage, so that the shell can hit the target faster when entering the laser guidance stage;

[0027] (4) The control method for accurately and quickly obtaining various parameters based on the multi-network model provided by the present invention makes the trajectory of the fired projectile smoother when entering the laser guidance stage, thereby avoiding the problem of missing the target due to insufficient yaw acceleration during the laser guidance process of the projectile. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A schematic flow chart of a control method for accurately and quickly obtaining various parameters based on a multi-network model according to a preferred embodiment of the present invention is shown;

[0029] Figure 2 A simulated ballistic trajectory diagram according to Example 1 of the present invention is shown. DETAILED DESCRIPTION

[0030] The present invention will be further described in detail below through the accompanying drawings and embodiments. Through these descriptions, the characteristics and advantages of the present invention will become more clear and distinct.

[0031] The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise noted.

[0032] According to a control method based on a multi-network model for accurate and rapid acquisition of various parameters provided by the present invention, Figure 1 As shown, the following steps are included:

[0033] S1. Setting a neural network model for calculating various elements to obtain shooting elements in a windless state, wherein the shooting elements include sight, loading and trajectory height;

[0034] S2, setting up a longitudinal wind neural network model, and obtaining the average longitudinal wind during the flight of the artillery shell according to the sight and trajectory height in the windless state shooting parameters, as well as the actual wind speed and wind direction, as the average trajectory longitudinal wind;

[0035] S3, setting up a neural network model for correction of various elements, taking the average longitudinal wind of the trajectory as input, outputting the correction value of the sight, correcting various elements of shooting in a windless state, and obtaining the final shooting elements;

[0036] S4. Adjust the launch platform according to the final shooting parameters and launch the shells.

[0037] In S1, the air pressure and temperature in the actual weather are fitted, and the fitting parameters and expected range are used as inputs of the neural network model for solving various parameters.

[0038] In the present invention, by adding the air pressure and temperature in the actual meteorological environment during the acquisition of the shooting parameters, the obtained shooting parameters can be closer to the actual use environment, so that the landing point of the shell is close to the target during laser guidance, so that the shell can hit the target faster when entering the laser guidance stage.

[0039] In a preferred embodiment, the fitting refers to fitting to a curve, and the fitting parameters refer to all parameters of the curve obtained by fitting.

[0040] According to the present invention, those skilled in the art can freely select the form of the fitting curve based on experience, such as an exponential curve, a quadratic curve, a cubic curve, etc.

[0041] In a preferred embodiment, the pressure fitting is performed using an exponential curve, which is expressed as where f p (x) represents the air pressure after fitting, x is the independent variable, a1 and a2 are the fitting parameters corresponding to the air pressure, and the temperature fitting adopts cubic function fitting, which is expressed as f t (x) = a3x 3 +a4x 2 +a5x+a6, where f t (x) represents the temperature after fitting, x is the independent variable, and a3, a4, a5, and a6 are the fitting parameters corresponding to the temperature. The above fitting curve not only has a high fitting accuracy, but also has a moderate amount of corresponding fitting parameters. Using it as one of the input parameters of the neural network model for solving various parameters can obtain more accurate predictions.

[0042] Based on a large amount of practical usage experience, the inventors found that the air pressure and temperature in actual meteorology are periodic and variable, and it is difficult to fit the air pressure and temperature of all years and different times into a curve. In a preferred embodiment, the year is divided into multiple parts according to the time approximation, for example, the year is divided into 12 parts by month, and preferably, the year is divided into 4 parts by quarter, and the air pressure and temperature in each time period are fit separately to obtain multiple fitting parameters.

[0043] Furthermore, for each fitting parameter, the neural network model for solving various parameters is trained independently. In actual use, the neural network model for solving various parameters trained with the fitting parameters corresponding to the current time is used.

[0044] In the present invention, there is no particular limitation on the specific network structure of the neural network model for solving various parameters, and those skilled in the art can freely select it based on experience.

[0045] In a preferred embodiment, the neural network model for solving various parameters adopts a BP neural network. More preferably, since the neural network model for solving various parameters only needs to perform a single prediction of shooting parameters, the neural network only uses two fully connected layers as hidden layers, each with 10 neurons, which not only ensures the prediction accuracy but also reduces the amount of calculation, so that the neural network model can be directly deployed in the fire control computer and achieve fast result output.

[0046] In a preferred embodiment, samples for training the neural network model for solving various parameters are obtained through ballistic trajectory simulation.

[0047] In the ballistic trajectory simulation, different air pressures, temperatures and expected ranges are set as simulation conditions. The ballistic trajectory of the projectile is obtained through simulation, and the sight, range and ballistic height of the projectile are obtained based on the ballistic trajectory.

[0048] Further preferably, in the ballistic trajectory simulation process, a nonlinear six-degree-of-freedom model is used to describe the ballistic trajectory. More preferably, the nonlinear six-degree-of-freedom model is expressed as:

[0049]

[0050]

[0051]

[0052]

[0053] Where m is the real-time mass of the projectile, P is the thrust of the projectile, X, Y, Z are the resistance, lift and lateral force of the projectile, α and β are the angle of attack and sideslip angle, respectively, and J x , J y, J z are the moments of inertia of the projectile along the three axes in the projectile system, M x 、M y 、M z are the moments of the projectile along the three axes in the projectile system, V m ,θ,ψ v Respectively represent the projectile speed, ballistic inclination and ballistic deviation; w x ,w y ,w z represents the angular velocity component of the velocity in the projectile coordinate system; x, y, z represent the three position components of the projectile in the ground coordinate system; ψ,γ represent the pitch angle, yaw angle and roll angle.

[0054] The geometric relationship between the angle of attack and the sideslip angle is as follows:

[0055]

[0056] Furthermore, the initial ballistic inclination angle θ0 obtained by simulation is the sight.

[0057] During the simulation, air temperature and air pressure affect the Mach number, which in turn affects the drag, thrust, and lateral force, causing the trajectory of the projectile to change.

[0058] According to the present invention, preferably, the neural network model for calculating various elements is only for one type of shells, and needs to be retrained when other types of shells are used.

[0059] According to a preferred embodiment of the present invention, in S2, according to the wind speed, wind direction, gun position reference and projectile-target distance in the actual weather, the longitudinal wind force magnitude of the projectile at different height layers is obtained, that is, the longitudinal wind array, which is expressed as [L h1 , …, L hi , …, L hn ], where hi represents different altitude layers, L hi It indicates the longitudinal wind size corresponding to the altitude layer, and hn indicates the total number of altitude layers;

[0060] The longitudinal wind neural network model is set up, and the sight and trajectory height in the windless state are used as input, and the longitudinal wind weight matrix of different altitude layers is output, which is expressed as [ω h1 ,…,ω hi ,…,ω hn ], where hi represents different altitude layers, ω hi Indicates the longitudinal wind weight of the altitude layer;

[0061] According to the longitudinal wind array and the longitudinal wind weight matrix, the average ballistic longitudinal wind L is obtained. w , expressed as:

[0062]

[0063] In a preferred embodiment, 200 meters is taken as one altitude layer to obtain longitudinal wind arrays and longitudinal wind weight matrices at different altitudes.

[0064] In a preferred embodiment, the longitudinal wind neural network model adopts a BP neural network. More preferably, since the longitudinal wind neural network model only needs to perform a single longitudinal wind weight matrix prediction, the neural network only uses two fully connected layers as hidden layers, each with 10 neurons, which not only ensures the prediction accuracy but also reduces the amount of calculation, so that the neural network model can be directly deployed in the fire control computer and achieve fast result output.

[0065] In a preferred embodiment, samples for training the longitudinal wind neural network model are obtained through ballistic simulation.

[0066] Preferably, in the trajectory simulation, different sights and trajectory heights are set for the projectiles, the longitudinal winds at different positions of the trajectory are obtained through simulation calculations, and then the weight matrices of the longitudinal winds at different altitudes are obtained through integration.

[0067] According to the present invention, preferably, the longitudinal wind neural network model is only for one type of artillery shells, and retraining is required when other types of artillery shells are needed.

[0068] In S3, the sight output by the neural network model for solving various elements is corrected by the sight correction amount to obtain a corrected sight, and the corrected sight and the loading and trajectory height output by the neural network model for solving various elements are used as the final shooting elements. In S3, preferably, the neural network model for correcting various elements adopts a BP neural network, and more preferably, since the neural network model for correcting various elements only needs to perform a single prediction of the sight correction amount, the neural network only uses two fully connected layers as hidden layers, each layer having 10 neurons, which not only ensures the prediction accuracy but also reduces the amount of calculation, so that the neural network model can be directly deployed in the fire control computer and achieve fast result output.

[0069] In a preferred embodiment, samples for training the various-element modified neural network model are obtained through a trajectory simulation model.

[0070] Preferably, in the trajectory simulation model, different sights are set in the windless state to obtain the trajectory landing point in the windless state; the trajectory longitudinal wind is increased, and the sight is corrected so that the obtained trajectory landing point is the same as the trajectory landing point in the windless state, and the sight, trajectory longitudinal wind and sight correction amount in the windless state are used as training samples.

[0071] According to the present invention, preferably, the various-element correction neural network model is only for one type of shells, and needs to be retrained when other types of shells are used.

[0072] In S4, the corrected firing parameters are used to launch the shells. Since the firing parameters are closer to the weather at the time of launch, the overlap between the shells and the preset trajectory is higher, so that when the shells enter the laser guidance, the landing point is close to the target position, so that the shells can hit the target faster when entering the laser guidance stage. In addition, the trajectory of the shells is smoother when entering the laser guidance stage, avoiding the problem of missing the target due to insufficient yaw acceleration during the laser guidance process.

[0073] Example

[0074] Example 1

[0075] A simulation experiment was carried out. The initial velocity was selected as 585 m / s, and the other simulation conditions were shown in Table 1.

[0076] Table 1 Simulation conditions

[0077]

[0078] The shooting parameters are obtained by:

[0079] The following steps are involved:

[0080] S1. Setting a neural network model for calculating various elements to obtain shooting elements in a windless state, wherein the shooting elements include sight, loading and trajectory height;

[0081] S2, setting up a longitudinal wind neural network model, and obtaining the average longitudinal wind during the flight of the artillery shell according to the sight and trajectory height in the windless state shooting parameters, as well as the actual wind speed and wind direction, as the average trajectory longitudinal wind;

[0082] S3, setting up a neural network model for correction of various elements, taking the average longitudinal wind of the trajectory as input, outputting the correction value of the sight, correcting various elements of shooting in a windless state, and obtaining the final shooting elements;

[0083] S4. Adjust the launch platform according to the final shooting parameters and launch the shells.

[0084] In S1, the air pressure and temperature in the actual weather are fitted, and the fitting parameters and expected range are used as inputs of the neural network model for solving various parameters. The air pressure is fitted by an exponential curve, and the temperature is fitted by a cubic function.

[0085] The various-element solving neural network model, the longitudinal wind neural network model and the various-element correcting neural network model all adopt BP neural network, use two fully connected layers as hidden layers, and each layer has 10 neurons.

[0086] In S2, according to the wind speed, wind direction, gun position reference and projectile-target distance in the actual weather, the longitudinal wind force on the shells at different altitudes is obtained, i.e., the longitudinal wind array.

[0087] Set up a longitudinal wind neural network model, take the sight and trajectory height in the windless state shooting parameters as input, and output the longitudinal wind weight matrix at different altitudes;

[0088] The average ballistic longitudinal wind is obtained based on the longitudinal wind array and the longitudinal wind weight matrix.

[0089] Taking 200 meters as a height layer, the longitudinal wind array and longitudinal wind weight matrix at different heights are obtained.

[0090] In S3, the sight output by the neural network model for all-element solution is corrected with the sight correction amount to obtain a corrected sight, and the corrected sight and the loading and trajectory height output by the neural network model for all-element solution are used as the final shooting elements.

[0091] According to the final shooting parameters, the ballistic trajectory results obtained by simulation are as follows Figure 2 As shown, from Figure 2 It can be seen that the output range of the trajectory is close to the expected range, and the specific error and accuracy are shown in Table 2.

[0092] Table 2 Error and accuracy

[0093]

[0094] It can be seen from Table 2 that the error values ​​under the three groups of expected ranges are around 20m, and the error rate is around 0.1%. It can be seen from Table 2 that the accuracy of the shooting parameters obtained in Example 1 is relatively high.

[0095] In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "front", "rear", etc. indicate positions or positional relationships based on the working state of the present invention, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", "third", and "fourth" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0096] The present invention has been described above in conjunction with preferred embodiments, but these embodiments are only exemplary and serve only as an illustration. On this basis, the present invention may be subjected to a variety of substitutions and improvements, all of which fall within the scope of protection of the present invention.

Claims

1. A control method for accurately and quickly obtaining parameters based on a multi-network model, characterized in that: The following steps are involved: S1. Setting a neural network model for calculating various elements to obtain shooting elements in a windless state, wherein the shooting elements include sight, loading and trajectory height; S2, setting up a longitudinal wind neural network model, and obtaining the average longitudinal wind during the flight of the artillery shell according to the sight and trajectory height in the windless state shooting parameters, as well as the actual wind speed and wind direction, as the average trajectory longitudinal wind; S3, setting up a neural network model for correction of various elements, taking the average longitudinal wind of the trajectory as input, outputting the correction value of the sight, correcting various elements of shooting in a windless state, and obtaining the final shooting elements; S4, adjusting the launch platform according to the final shooting elements to launch the shells; In S1, the air pressure and temperature in the actual weather are fitted, and the fitting parameters and the expected range are used as inputs of the neural network model for solving various parameters; In S2, according to the wind speed, wind direction, gun position reference and projectile-target distance in the actual weather, the longitudinal wind force on the shells at different altitudes is obtained, i.e., the longitudinal wind array. Set up a longitudinal wind neural network model, take the sight and trajectory height in the windless state shooting parameters as input, and output the longitudinal wind weight matrix at different altitudes; The average ballistic longitudinal wind is obtained based on the longitudinal wind array and the longitudinal wind weight matrix.

2. The control method for accurately and quickly obtaining parameters based on a multi-network model according to claim 1 is characterized in that: The air pressure fitting is performed using an exponential curve. The temperature fitting adopts cubic function fitting.

3. The control method for accurate and rapid acquisition of parameters based on a multi-network model according to claim 1 is characterized in that: The year is divided into multiple parts, and the air pressure and temperature in each time period are fitted separately to obtain multiple fitting parameters. For each fitting parameter, the neural network model for solving the various parameters is trained independently.

4. The control method for accurately and quickly obtaining parameters based on a multi-network model according to claim 3 is characterized in that: The year is divided into four parts according to quarters. In actual use, the neural network model with various parameters trained with fitting parameters corresponding to the quarter of the current time is used for solution.

5. The control method for accurately and quickly obtaining parameters based on a multi-network model according to claim 1 is characterized in that: The neural network model for solving various parameters adopts BP neural network, uses two fully connected layers as hidden layers, and each layer has 10 neurons.

6. The control method for accurately and quickly obtaining parameters based on a multi-network model according to claim 1 is characterized in that: The samples for training the neural network model for solving various parameters are obtained through ballistic trajectory simulation.

7. The control method for accurately and quickly obtaining parameters based on a multi-network model according to claim 1 is characterized in that: Taking 200 meters as a height layer, the longitudinal wind array and longitudinal wind weight matrix at different heights are obtained.

8. The control method for accurately and quickly obtaining parameters based on a multi-network model according to claim 1 is characterized in that: In S3, the sight output by the neural network model for all-element solution is corrected with the sight correction amount to obtain a corrected sight, and the corrected sight and the loading and trajectory height output by the neural network model for all-element solution are used as the final shooting elements.

Images