Calculation method of initial parameters of spacecraft standard trajectory based on BP neural network

Abstract

The present invention discloses a method for calculating the initial parameters of the standard trajectory of a spacecraft based on a BP neural network. First, an initial sample is generated by iterative integration of the standard trajectory. Then, a neural network model of the aiming azimuth angle and a neural network model of the shutdown time are established respectively through a supervised learning algorithm. The input layers are the launch point latitude B0, the launch point elevation H0, and the geodetic distance L between the launch point and the landing point. d and the Dadi azimuth A d , the output layer corresponds to the aiming azimuth A mz and shutdown time t k The simulation results show that the longitudinal deviation of the aircraft landing point predicted by the neural network is less than 10m, and the lateral deviation is less than 20m, which meets the accuracy requirements. At the same time, compared with the traditional integration method, the neural network only takes an average of 0.014857s to complete the calculation of the initial parameters, which shortens the calculation time by 29,000 times and can achieve the goal of rapid launch.

Description

Technical Field

[0001] The present invention relates to the technical field of spacecraft, and in particular to a method for calculating initial elements of a spacecraft standard trajectory based on a BP neural network. Background Art

[0002] Parameter calculation is the process of precalculating the parameters that ensure an aircraft follows a predetermined standard trajectory toward its target, based on launch conditions and aircraft performance parameters. This allows for flight control and precision control. Parameter calculation is a crucial technical support for the operational use of aircraft weapons. It directly impacts flight control stability and firing accuracy, and its accuracy and timeliness are crucial to the success or failure of aircraft weapon launches.

[0003] With the improvement of combat requirements, the requirements for rapid launch of aircraft are becoming increasingly higher, and the launch preparation time is greatly shortened, which has led to the requirement to shorten the calculation time of various parameters.

[0004] Current methods for calculating various parameters include shot table fitting, free-segment analytical solutions, and integral calculations. The shot table fitting method requires a large amount of pre-calculated data as a basis for fitting, which is computationally intensive and requires the creation of complex data tables. Multiple trajectory iterations are required to obtain the initial parameters, resulting in a long calculation time. The free-segment analytical solution method is faster, but the polynomial form after fitting is complex, and the accuracy of the polynomial approximation calculation is limited. For many years, integral calculation methods have remained the mainstream method for calculating various parameters. The integral calculation method first calculates the geodetic distance and geodetic azimuth from the launch point to the impact point, using these initial values ​​to roughly calculate the active segment shutdown time. Using this rough shutdown time and geodetic azimuth, multiple trajectory iterations are required to accurately calculate the shutdown time and aiming azimuth of the standard trajectory. The integral calculation method suffers from complex integral solutions and slow iterative solutions, making it increasingly unable to meet the needs of new launch methods such as rapid maneuverable launch.

[0005] Building on the traditional iterative method, Wang Minghai et al. proposed a fast iterative method that decomposes the binary iterations of the impact point range deviation and azimuth deviation into two single-variable iterations. This method reduces the number of trajectory iterations, but trajectory iterations are still required. Yong Enmi et al. established a mathematical model for calculating the standard trajectory shutdown time, launch azimuth, shutdown equation, and guidance equation coefficients, and designed and implemented software for calculating the various parameters of the aircraft launch based on this model. However, this method still fails to completely resolve the problems of complex integral calculations, iterative loops that can occur in certain situations, and the inability to quantify the strong coupling between the aiming azimuth and shutdown time, thus failing to achieve the goal of rapidly calculating the initial parameters. Summary of the Invention

[0006] In view of the above problems, the present invention aims to provide a method for calculating the initial parameters of the standard trajectory of a spacecraft based on a BP neural network, which can greatly shorten the time of parameter calculation while ensuring the calculation accuracy.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] The method for calculating initial parameters of the standard trajectory of a spacecraft based on BP neural network is characterized by comprising the following steps:

[0009] S1: Analyze and determine the factors that affect the calculation of initial parameters;

[0010] S2: Based on the factors that affect the initial calculation of various parameters, the aiming azimuth angle A is generated through the standard trajectory integration iterative method mz and shutdown time t k The initial sample of

[0011] S3: Using the initial sample in step S2, an initial element calculation model based on the BP neural network is established;

[0012] S4: Calculate the initial parameters of the standard trajectory of the spacecraft using the initial parameter calculation model established in step S3.

[0013] Furthermore, the specific operation of step S2 includes the following steps:

[0014] S201: Calculate the earth distance L between the launch point and the target point based on the given launch point and target point m and the Dadi azimuth A m ;

[0015] S202: Based on the earth distance L between the launch point and the target point m and the Dadi azimuth A m Estimated shutdown time and initial value of aiming azimuth t k0 、A mz0 ;

[0016] S203: The estimated shutdown time and the initial value of the aiming azimuth angle t k0 、A mz0 Substitute the standard trajectory to obtain the landing point of the aircraft;

[0017] S204: Calculate the earth distance L between the launch point and the landing point of the aircraft d and the Dadi azimuth A d ;

[0018] S205: Based on the earth distance L between the launch point and the landing point d and the Dadi azimuth A d, solve the initial shutdown time t by iterative method k and aiming azimuth A mz .

[0019] Furthermore, the specific operation of step S205 includes the following steps:

[0020] S2051: Express the relationship between the geodetic distance between the launch point and the landing point, the geodetic azimuth, the aiming azimuth, and the shutdown time.

[0021]

[0022] At the shutdown time and the initial value of the aiming azimuth (t k0 , A mz0 ) to obtain

[0023]

[0024] Where A() and L() are implicit function expressions, which respectively represent the calculation of geodetic azimuth and geodetic distance from the shutdown time and aiming azimuth; ΔL and ΔA are the geodetic distance and geodetic azimuth difference between the target point and the landing point of the aircraft, respectively.

[0025] S2052: Δt obtained by solving step S1051 k and ΔA mz Perform iteration, then

[0026]

[0027] Where k and k+1 represent the number of iterations;

[0028] S2053: The shutdown time t obtained in step S2052 is k (k+1) and aiming azimuth A mz Substitute (k+1) into the standard trajectory to calculate the new landing point of the aircraft, and repeat steps S204, S2051 and S2052 until the geodetic distance and geodetic azimuth difference ΔL and ΔA between the new landing point and the target point meet the accuracy requirements, and the initial parameters t are obtained. k and A mz .

[0029] Furthermore, the initial element calculation model described in step S3 includes an aiming azimuth angle neural network model and a shutdown time neural network model.

[0030] Furthermore, the aiming azimuth neural network model in step S3 includes an input layer, two hidden layers and an output layer. The number of nodes in the input layer is 4, and the input quantities are the launch point latitude B0, the launch point elevation H0, the geodetic distance L between the launch point and the landing point, and the target azimuth angle. d and the Dadi azimuth Ad The number of nodes in the two hidden layers is 19 and 18 respectively, and the activation function of the two hidden layers is Sigmoid. The activation function of the output layer is the linear function Purline, and the output is the aiming azimuth A. mz ,

[0031] Furthermore, the shutdown time neural network model in step S3 includes an input layer, two hidden layers and an output layer. The number of nodes in the input layer is 4, and the input quantities are the latitude B0 of the launch point, the altitude H0 of the launch point, the earth distance L between the launch point and the landing point, and the distance L between the launch point and the landing point. d and the Dadi azimuth A d The number of nodes in the two hidden layers is 15 and 14 respectively, and both hidden layers are activated by the logarithmic Sigmoid function. The activation function of the output layer is the linear function Purline, and the output is the shutdown time t k .

[0032] Furthermore, the specific operation of step S3 includes the following steps:

[0033] S301: normalize the initial samples obtained in step S2;

[0034] S302: establishing an aiming azimuth angle neural network model and a shutdown time neural network model based on a BP neural network respectively;

[0035] S303: Perform network training and testing on the aiming azimuth neural network model and the shutdown time neural network model.

[0036] Furthermore, the specific method of normalization processing in step S301 is:

[0037]

[0038] In the formula, y is the normalized input sample data, y max is the upper limit of the normalized interval, y min is the lower limit of the normalized interval, x is the input sample data to be normalized, x min is the minimum value of the input sample data, x max is the maximum value of the input sample data.

[0039] The beneficial effects of the present invention are:

[0040] The method for calculating the initial parameters of the standard trajectory of a spacecraft based on the BP neural network in the present invention first generates an initial sample by iteratively integrating the standard trajectory, and then establishes a neural network model for the aiming azimuth angle and a neural network model for the shutdown time respectively through a supervised learning algorithm. The input layers are the launch point latitude B0, the launch point elevation H0, and the geodetic distance L between the launch point and the landing point.d and the Dadi azimuth A d , the output layer corresponds to the aiming azimuth A mz and shutdown time t k The simulation results show that the longitudinal deviation of the aircraft landing point predicted by the neural network is less than 10m, and the lateral deviation is less than 20m, which meets the accuracy requirements. At the same time, compared with the traditional integration method, the neural network only takes an average of 0.014857s to complete the calculation of the initial parameters, which shortens the calculation time by 29,000 times and can achieve the goal of rapid launch. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 These are factors that affect the calculation of initial parameters in the present invention.

[0042] Figure 2 Flowchart of solving initial variables for traditional methods.

[0043] Figure 3 The flowchart of the process of establishing the initial neural network model in the present invention.

[0044] Figure 4 This is the neural network structure for aiming at the azimuth angle in the present invention.

[0045] Figure 5 This is the neural network structure of the shutdown time in the present invention.

[0046] Figure 6 It is the convergence curve of the error in the training process of the neural network aiming at the azimuth angle in the present invention.

[0047] Figure 7 This is the neural network error convergence curve of the shutdown time in the present invention.

[0048] Figure 8 It is the neural network prediction error of the aiming azimuth in the simulation experiment of the present invention.

[0049] Figure 9 It is the neural network prediction error of shutdown time in the simulation experiment of the present invention. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0051] The method for calculating the initial parameters of the standard trajectory of a spacecraft based on BP neural network includes the following steps:

[0052] S1: Analyze and determine the factors that affect the calculation of initial parameters;

[0053] Specifically, the starting data of the standard trajectory are the launch point longitude L0, the launch point latitude B0, the launch point elevation H0, the target point longitude L m 、Latitude of target point B m , Target point elevation H m In the force analysis of an actual aircraft flight, there are relatively few forces correlated with geodetic longitude. Because the Earth is a rotating ellipsoid, longitude has almost no effect on the magnitude of the force acting on the aircraft, and the effect of longitude on the standard trajectory calculation can be ignored. In the zonal and eustatic directions, the unevenness of the Earth's radius significantly affects the flight of the aircraft, affecting the calculation of the aircraft's initial parameters. The magnitude of gravity is closely related to the elevation of the launch point. After multiple simulation analyses, it has been shown that elevation has a significant impact on the calculation of initial parameters. Assuming the target point elevation is consistent, only the influence of the launch point elevation H0 on the calculation of initial parameters is considered.

[0054] (1)Latitude of the launch point

[0055] Let the true value of the latitude of the launch point be The measurement deviation is ΔB0, then the measured value B0 of the launch point latitude is

[0056]

[0057] Therefore, the standard trajectory is When point is used as the launch point, the actual launch point is Where, and are the true values ​​of the longitude and elevation of the launch point, respectively;

[0058] The aircraft landing point deviation caused by latitude deviation is

[0059] ΔL B =ΔL B，j (ΔB0)+ΔL B，F (ΔB0) (2)

[0060] Where ΔL B ΔL is the aircraft landing point deviation, including the lateral deviation and longitudinal deviation of the aircraft landing point; B，j is the geometric effect, ΔL B，F It is a dynamic effect caused by the deviation of gravity and inertial force;

[0061] According to the knowledge of geometry, we can get

[0062]

[0063] is the mean radius of the Earth;

[0064] Gravity calculation formula for aircraft

[0065]

[0066] From formulas (4) and (5), we can see that when the actual launch point latitude of the aircraft deviates from the latitude calculated by various parameters by ΔB0, the earth's gravity produces a corresponding deviation Δg. The real-time flight speed and position of the aircraft are related to the magnitude of gravity. The deviation of the aircraft's position and speed can be expressed as

[0067]

[0068] Where g x is the projection of gravitational acceleration g in the X direction of the inertial coordinate system, g y is the projection of gravitational acceleration g in the Y direction of the inertial coordinate system; g z is the projection of gravitational acceleration g in the Z direction of the inertial coordinate system, g r is the radial gravity of the gravitational acceleration g at the orbital radius, x is the projection of the real-time position of the aircraft on the X-axis of the inertial system, y is the projection of the real-time position of the aircraft on the Y-axis of the inertial system, z is the projection of the real-time position of the aircraft on the Z-axis of the inertial system, r is the distance between the real-time position of the aircraft and the center of the earth, w is the angular velocity of the earth's rotation, R ox is the X-axis projection of the launch point’s center of mass on the inertial system, R oy is the Y-axis projection of the launch point’s center of mass on the inertial system, R oz is the Z-axis projection of the launch point's center of mass on the inertial system, g w is the gravitational acceleration g perpendicular to the orbital radius, w x is the X-axis projection of the Earth's rotational angular velocity in the inertial system, w y is the Y-axis projection of the Earth's rotational angular velocity in the inertial system, w z is the Z-axis projection of the Earth's rotational angular velocity in the inertial system, A mz is the aiming azimuth; is the deviation of the aircraft position, is the deviation of the aircraft speed;

[0069] From formula (6), we can see that the range and speed of the aircraft are functions of the latitude of the launch point. When the latitude changes, the position and speed of the actual trajectory deviate from the standard trajectory.

[0070] ΔL B，F (ΔB0) is affected by the inertial force deviation, and has the following effects on the inertial acceleration:

[0071]

[0072] For the Coriolis inertial acceleration,

[0073]

[0074] Where, is the projection of the acceleration in the X direction of the inertial coordinate system, is the projection of the acceleration in the Y direction of the inertial coordinate system, is the projection of the acceleration in the Z direction of the inertial coordinate system; r x is the projection of the real-time center-of-the-earth radius of the aircraft in the X direction of the inertial coordinate system, r y is the projection of the real-time center-of-the-earth radius of the aircraft in the Y direction of the inertial coordinate system, r z It is the projection of the real-time Earth-centered radius of the aircraft in the Z direction of the inertial coordinate system; is the projection of the Coriolis acceleration in the X direction of the inertial coordinate system, is the projection of the Coriolis acceleration in the Y direction of the inertial coordinate system, is the projection of the Coriolis acceleration in the Z direction of the inertial coordinate system;

[0075] From formula (5), we can see that w x , w y , w z The latitude B0 is related to the acceleration generated by the drag inertial force. Formulas (7) and (8) show that the acceleration generated by the drag inertial force is related to the latitude B0. Formulas (9) and (10) show that the acceleration generated by the Coriolis inertial force is related to the latitude B0. When the latitude deviates by ΔB0, the drag acceleration and the Coriolis acceleration will inevitably deviate, affecting the accuracy of the aircraft's landing point.

[0076] For the same range and direction, changes in the launch point's latitude can cause significant deviations in the impact point. When the geodetic azimuth is 0°, a 1° change in latitude results in a vertical deviation of approximately 1 to 3 meters and a lateral deviation of approximately 15 to 18 meters.

[0077] (2) Elevation of the launch point

[0078] The launch point elevation is the distance from the launch point to the geoid along a vertical line, typically measured using theodolite trigonometric height measurement. Launch point elevation deviation significantly affects the trajectory of the active phase of the aircraft, primarily affecting the x- and y-direction position and velocity of the aircraft's shutdown point during the active phase of flight. This effect is nonlinear.

[0079] The launch point elevation deviation ΔH0 refers to the difference between the actual sea level altitude of the launch point and the measured sea level altitude; the launch point elevation deviation ΔL H (ΔH0) can be divided into the deviation ΔL caused by geometric factors H，j (ΔH0) and the deviation ΔL caused by dynamics H,F (ΔH0).

[0080] The size of the geometric deviation is

[0081]

[0082] Where θ is the trajectory inclination angle of the vehicle re-entering the atmosphere, and β is the range angle;

[0083] According to the mathematical model, we can get

[0084]

[0085] From equations (4), (12), (13), and (14), it can be seen that the magnitude of gravity is closely related to the launch point elevation. When the launch point elevation deviates by ΔH0, a gravity deviation Δg will occur. This gravity deviation will inevitably cause position deviation, which in turn causes deviations in acceleration, aerodynamic force, and engine thrust. The difference in force further exacerbates the deviation in position and velocity, forming a divergent propagation of error, which has a significant impact on the trajectory landing accuracy.

[0086] (3) Geodetic distance and geodetic azimuth

[0087] Spacecraft trajectory calculations include geodetic element calculations and firing element calculations. Geodetic element calculations use the geocentric coordinates of the launch and target points, employing the inverse Bessel or Baumann geodetic problem formulas, to determine the geodetic distance and geodetic azimuth from the launch point to the target. This provides the baseline data for firing element calculations. Geodetic element calculations are the foundation of all element calculations and significantly impact the accuracy of the spacecraft's landing point.

[0088] The standard trajectory under the given launch point conditions, its landing point is a function of the aiming azimuth and the shutdown time. The large azimuth angle A between the launch point and the landing point of the aircraft d Distance from the earth L d It is also determined by the aiming azimuth and the shutdown time. The relationship between the various parameters is shown in formula (15).

[0089]

[0090] In summary, the standard trajectory calculation data in the present invention is converted into: the launch point latitude B0, the launch point elevation H0, the geodetic distance between the launch point and the target point L d and the azimuth angle A between the launch point and the target point d , factors that affect the calculation of initial parameters are as follows Figure 1 shown.

[0091] Furthermore, S2: Based on the factors affecting the initial calculation of various parameters, the aiming azimuth angle A is generated by the standard trajectory integration iterative method. mz and shutdown time t k The initial sample of

[0092] The traditional method of calculating the initial parameters is to solve it by iterative calculation given the launch point and target point information. The process of solving the standard trajectory aiming azimuth and shutdown time is as follows: Figure 2 Specifically,

[0093] S201: Calculate the earth distance L between the launch point and the target point based on the given launch point and target point m and the Dadi azimuth A m ;

[0094] S202: Based on the earth distance L between the launch point and the target point m and the Dadi azimuth A m Estimated shutdown time and initial value of aiming azimuth t k0 、A mz0 ;

[0095] S203: The estimated shutdown time and the initial value of the aiming azimuth angle t k0 、A mz0 Substitute the standard trajectory to obtain the landing point of the aircraft;

[0096] S204: Calculate the earth distance L between the launch point and the landing point of the aircraft d and the Dadi azimuth A d ;

[0097] S205: Based on the earth distance L between the launch point and the landing point d and the Dadi azimuth A d , solve the initial shutdown time t by iterative method k and aiming azimuth A mz .

[0098] Formula (15) is used to calculate the initial value of the aiming azimuth angle (t k0 , A mz0 ) to obtain

[0099]

[0100] Where A() and L() are implicit function expressions, which respectively represent the calculation of geodetic azimuth and geodetic distance from the shutdown time and aiming azimuth; ΔL and ΔA are the geodetic distance and geodetic azimuth difference between the target point and the landing point of the aircraft, respectively.

[0101] According to the obtained Δt k and ΔA mz Iterate and let

[0102]

[0103] Where k and k+1 represent the number of iterations;

[0104] The shutdown time t k (k+1) and aiming azimuth A mz Substitute (k+1) into the standard trajectory to calculate the new landing point of the aircraft, and find the ΔL and ΔA between the new landing point and the target point. When ΔL and ΔA meet the accuracy requirements, the initial parameters t are obtained. k and A mz Otherwise, repeat the iterative calculation according to the above method until ΔL and ΔA meet the accuracy requirements.

[0105] Further, S3: using the initial sample in step S2 to establish an initial element calculation model based on the BP neural network;

[0106] Aiming azimuth and shutdown time are two important parameters for calculating the standard trajectory, and are affected by the launch point latitude, launch point elevation, geodetic azimuth and geodetic distance at the same time. After experimental analysis, a neural network model is established with the launch point latitude, launch point elevation, geodetic azimuth and geodetic distance as input, and the aiming azimuth and shutdown time as output. The nonlinear mapping relationship between the four inputs and the two outputs is fitted in parallel, and the effect of neural network training is not ideal. Therefore, two neural network models with different structures are established in the present invention, which respectively fit the nonlinear mapping relationship between the launch point latitude, launch point elevation, geodetic azimuth and geodetic distance and the aiming azimuth, and the nonlinear mapping relationship between the launch point latitude, launch point elevation, geodetic azimuth and geodetic distance and the shutdown time, thereby realizing the rapid calculation of the aiming azimuth and shutdown time. The two neural network models include an aiming azimuth neural network model and a shutdown time neural network model. The process of establishing the initial neural network model is shown in the attached figure. Figure 3 shown.

[0107] Specifically, the aiming azimuth neural network model is as shown in the attached Figure 4 As shown, it includes an input layer, two hidden layers and an output layer. The number of nodes in the input layer is 4, and the input quantities are the launch point latitude B0, the launch point elevation H0, and the geodetic distance L between the launch point and the landing point. d and the Dadi azimuth A d The number of nodes in the two hidden layers is 19 and 18 respectively, and the activation function of the two hidden layers is Sigmoid. The activation function of the output layer is the linear function Purline, and the output is the aiming azimuth A. mz .

[0108] The shutdown time neural network model is as shown in the attached Figure 5 As shown, it includes an input layer, two hidden layers and an output layer. The number of nodes in the input layer is 4, and the input quantities are the launch point latitude B0, the launch point elevation H0, and the geodetic distance L between the launch point and the landing point. d and the Dadi azimuth Ad The number of nodes in the two hidden layers is 15 and 14 respectively, and both hidden layers are activated by the logarithmic Sigmoid function. The activation function of the output layer is the linear function Purline, and the output is the shutdown time t k .

[0109] More specifically, the process of establishing the aiming azimuth neural network model and the shutdown time neural network model includes the following steps:

[0110] S301: normalize the initial samples obtained in step S2;

[0111] Aiming azimuth A mz and shutdown time t k The factors affecting the launch point are the launch point latitude B0, the launch point elevation H0, the geodetic distance L between the launch point and the landing point d and the Dadi azimuth A d , these four factors are used as the input parameters of the neural network. The output parameters of the neural network are aiming azimuth A mz , shutdown time t k Calculate A for each combination by integrating formula (17) mz and t k , forming 46800 sets of standard samples. The sample generation conditions of the present invention are: B0 range is 30°~35°, H0 range is 300m~600m, A d The range is 0°~360°, L d The range is 200km to 600km. Gridding is used to generate samples within the above range, and the samples can cover all situations within the range.

[0112] There is a large gap in the order of magnitude between the sample data obtained from the four input variables, and the BP neural network is very sensitive to data changes. If the input data differs too much in order of magnitude, the weights of the neural network will also be affected by a large order of magnitude, which is not conducive to the neural network's learning of nonlinear mapping relationships.

[0113] In order to avoid errors in the neural network caused by differences in the magnitude of sample data, the four input data need to be dimensionless and controlled within a relatively close range. Therefore, the input data are normalized according to the following formula.

[0114]

[0115] In the formula, y is the normalized input sample data, y max is the upper limit of the normalized interval, y min is the lower limit of the normalized interval, x is the input sample data to be normalized, x min is the minimum value of the input sample data, xmax is the maximum value of the input sample data.

[0116] Record this normalized mapping and use it to denormalize the prediction results of the test samples. Shuffle the order of the 46,800 groups of standard samples, take the first 37,800 groups of samples as the training data of the neural network, and the last 9,000 groups of samples as the test data of the neural network.

[0117] S302: establishing an aiming azimuth angle neural network model and a shutdown time neural network model based on a BP neural network respectively;

[0118] According to the above-mentioned aiming azimuth angle neural network model and shutdown time neural network model structures, an aiming azimuth angle neural network model and a shutdown time neural network model are established respectively.

[0119] S303: Perform network training and testing on the aiming azimuth neural network model and the shutdown time neural network model.

[0120] The order of the 46,800 groups of standard samples was shuffled, and the first 37,800 groups of samples were taken as training data for two BP neural networks. The two BP neural networks were trained separately to obtain the weights and thresholds of the neural networks.

[0121] The two BP networks above adopt the same training method. The maximum number of neural network training times is 10000, the learning rate is 0.01, and the network target error is 1×10 -10 The training method is the Levenberg-Marquardt (LM) algorithm, and the loss function is the mean squared error (MSE). The LM algorithm is suitable for small and medium-sized neural networks and has the advantage of fast convergence speed. The training process of the two neural networks is shown in the attached figure. Figure 6 and attached Figure 7 shown.

[0122] Further, S4: using the initial element calculation model established in step S3 to calculate the initial elements of the standard trajectory of the trajectory aircraft.

[0123] The launch point latitude B0, the launch point elevation H0, the geodetic distance between the launch point and the landing point L d and the Dadi azimuth A d The data is input into the neural network trained in step S3, and the initial elements can be aimed at the azimuth angle A. mz and shutdown time t k Perform fast automatic calculations.

[0124] Simulation experiment:

[0125] In order to test the effectiveness of the trained BP neural network, the last 9000 sample data sets that did not participate in the training network were used to verify the effect of the neural network on the initial parameters calculation. The results of the neural network calculation were compared with the results of the traditional standard trajectory integral iterative calculation. The simulation results are shown in the attached figure. Figure 8 and 9 shown.

[0126] By the attached Figure 8 It can be seen that the absolute value of the maximum error between the aiming azimuth calculated by the neural network and the aiming azimuth calculated by the standard trajectory iteration is less than 0.0095 rad, and the mean square error of the error is 2.8752×10 - 5 rad. Figure 9 It can be seen that the absolute value of the maximum error between the shutdown time calculated by the neural network and the shutdown time calculated by the standard trajectory iteration is less than 0.005217s, and the mean square error of the error is 7.3619×10 -5 s, the prediction errors of aiming azimuth and shutdown time are shown in Table 1 below.

[0127] Table 1 Prediction errors of aiming azimuth and shutdown time

[0128]

[0129] The neural network's predictions of the shutdown time and aiming azimuth were evaluated using the longitudinal and lateral deviations of the landing point as evaluation criteria. Substituting the neural network's predicted aiming azimuth and shutdown time into the standard trajectory iterative integral, the lateral deviation ΔH∈(14, 20)m and the longitudinal deviation ΔL∈(2, 10)m of the landing point were calculated, demonstrating that the neural network model performs well in calculating the initial parameters.

[0130] The experimental environment is Windows 7 operating system, the memory space is 12GB, and the processor is Intel(R) Xeon(R) CPU E5-1603 0@2.80GHz.

[0131] The QueryPerformanceFrequency and QueryPerformanceCounter functions are used to calculate the time it takes to calculate the initial parameters. These functions return the frequency of a high-precision counter supported by the hardware. During standard trajectory initialization, the QueryPerformanceFrequency function is called to obtain the clock frequency of the computer's internal timer. The QueryPerformanceCounter function is called before and after the initial parameters calculation. The difference between the two counts and the clock frequency can be used to calculate the precise time it takes to calculate the initial parameters.

[0132] The average time consumption for calculating the initial parameters using the traditional iterative integration method and the neural network model is shown in Table 2.

[0133] Table 2 Comparison of average time consumption of two methods

[0134]

[0135] It can be seen from Table 2 that the neural network model proposed in this invention greatly improves the calculation efficiency of the initial parameters of the standard trajectory.

[0136] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for calculating initial parameters of a spacecraft standard trajectory based on a BP neural network, characterized in that: The following steps are included: S1: Analyze and determine the factors that affect the calculation of initial parameters; S2: Based on the factors that affect the initial calculation of various parameters, the aiming azimuth angle A is generated through the standard trajectory integration iterative method mz and shutdown time t k The initial sample of S3: Using the initial sample in step S2, an initial element calculation model based on the BP neural network is established; S4: Calculating the initial elements of the standard trajectory of the spacecraft using the initial element calculation model established in step S3; The specific operation of step S2 includes the following steps: S201: Calculate the earth distance L between the launch point and the target point based on the given launch point and target point m and the Dadi azimuth A m ; S202: Based on the earth distance L between the launch point and the target point m and the Dadi azimuth A m Estimated shutdown time and initial value of aiming azimuth t k0 、A mz0 ; S203: The estimated shutdown time and the initial value of the aiming azimuth angle t k0 、A mz0 Substitute the standard trajectory to obtain the landing point of the aircraft; S204: Calculate the earth distance L between the launch point and the landing point of the aircraft d and the Dadi azimuth A d ; S205: Based on the earth distance L between the launch point and the landing point d and the Dadi azimuth A d , solve the initial shutdown time t by iterative method k and aiming azimuth A mz ; The specific operation of step S205 includes the following steps: S2051: Express the relationship between the geodetic distance between the launch point and the landing point, the geodetic azimuth, the aiming azimuth, and the shutdown time. At the shutdown time and the initial value of the aiming azimuth (t k0 , A mz0 ) to obtain Where A() and L() are implicit function expressions, which respectively represent the calculation of geodetic azimuth and geodetic distance from the shutdown time and aiming azimuth; ΔL and ΔA are the geodetic distance and geodetic azimuth difference between the target point and the landing point of the aircraft, respectively. S2052: Δt obtained by solving step S1051 k and ΔA mz Perform iteration, then Where k and k+1 represent the number of iterations; S2053: The shutdown time t obtained in step S2052 is k (k+1) and aiming azimuth A mz Substitute (k+1) into the standard trajectory to calculate the new landing point of the aircraft, and repeat steps S204, S2051 and S2052 until the geodetic distance and geodetic azimuth difference ΔL and ΔA between the new landing point and the target point meet the accuracy requirements, and the initial parameters t are obtained. k and A mz .

2. The method for calculating initial parameters of a spacecraft standard trajectory based on a BP neural network according to claim 1, characterized in that: The initial parameter calculation model described in step S3 includes an aiming azimuth neural network model and a shutdown time neural network model.

3. The method for calculating initial parameters of a spacecraft standard trajectory based on a BP neural network according to claim 2, characterized in that: The aiming azimuth neural network model in step S3 includes an input layer, two hidden layers and an output layer. The number of nodes in the input layer is 4, and the input quantities are the launch point latitude B0, the launch point elevation H0, the earth distance L between the launch point and the landing point, and the target position. d and the Dadi azimuth A d The number of nodes in the two hidden layers is 19 and 18 respectively, and the activation function of the two hidden layers is Sigmoid. The activation function of the output layer is the linear function Purline, and the output is the aiming azimuth A. mz。 4. The method for calculating initial parameters of a spacecraft standard trajectory based on a BP neural network according to claim 2, characterized in that: The shutdown time neural network model in step S3 includes an input layer, two hidden layers and an output layer. The number of nodes in the input layer is 4, and the input quantities are the launch point latitude B0, the launch point elevation H0, the geodetic distance L between the launch point and the landing point. d and the Dadi azimuth A d The number of nodes in the two hidden layers is 15 and 14 respectively, and both hidden layers are activated by the logarithmic Sigmoid function. The activation function of the output layer is the linear function Purline, and the output is the shutdown time t k .

5. The method for calculating initial parameters of a spacecraft standard trajectory based on a BP neural network according to claim 2, characterized in that: The specific operation of step S3 includes the following steps: S301: normalize the initial samples obtained in step S2; S302: establishing an aiming azimuth angle neural network model and a shutdown time neural network model based on a BP neural network respectively; S303: Perform network training and testing on the aiming azimuth neural network model and the shutdown time neural network model.

6. The method for calculating initial parameters of a spacecraft standard trajectory based on a BP neural network according to claim 5, characterized in that: The specific method of normalization processing in step S301 is: In the formula, y is the normalized input sample data, y max is the upper limit of the normalized interval, y min is the lower limit of the normalized interval, x is the input sample data to be normalized, x min is the minimum value of the input sample data, x max is the maximum value of the input sample data.

Images