Rapid ballistic calculation method and system, computer equipment and storage medium

Abstract

The invention relates to the technical field of ballistic calculation, and discloses a rapid ballistic calculation method and system, computer equipment and a storage medium, and the method comprises the steps: generating ballistic calculation parameters based on rasterization, carrying out calculation, obtaining multiple pieces of original ballistic data, carrying out the feature extraction and segmentation processing of the original ballistic data, and obtaining a sample data set; the sample data set is adopted to train a trajectory resolving model, a trained trajectory resolving model is obtained, and the trajectory resolving model comprises a front section trajectory sub-model, a middle section trajectory sub-model, a rear section trajectory sub-model and a trajectory feature sub-model; and inputting to-be-calculated parameters into the trained trajectory calculation model for calculation to obtain trajectory scatter data. According to the method, the deep learning technology and the actual ballistic segmentation characteristics are combined, the rapid ballistic prediction method based on the segmentation training strategy is provided, the rapid ballistic calculation at the millisecond level is achieved on the premise that the prediction precision is guaranteed, and the requirements for missile combat mission planning and decision real-time performance are met.

Description

Technical Field

[0001] The present invention relates to the technical field of trajectory calculation, and in particular to a fast trajectory calculation method, system, computer equipment and storage medium. Background Art

[0002] The missile trajectory prediction problem can be divided into target missile trajectory prediction and our missile trajectory prediction. The difference between the two lies in the different starting points. Target missile trajectory prediction enables our missile interception system to track and predict the target missile flight trajectory as quickly as possible, derive the interception point position, and provide guidance for our interceptor missile flight trajectory planning, thereby improving the interception success rate. Our missile trajectory prediction mainly provides trajectory simulation verification for task allocation and decision-making results in missile combat.

[0003] At present, our missile trajectory calculation methods are mainly divided into two categories: one is the traditional analytical method and numerical integration method. The former derives the analytical solution of the trajectory of hypersonic vehicles through perturbation method and spectral decomposition method, and the numerical integration method (such as Runge-Kutta method) solves nonlinear differential equations through iteration. The other method is the trajectory prediction method based on neural network, which fits and approximates the nonlinear trajectory model through neural network to achieve rapid trajectory prediction.

[0004] Although the existing missile trajectory prediction technology has made great progress, it still has some obvious shortcomings in practical applications. Among them, although the analytical method can derive the theoretical ballistic analytical solution, its complex derivation process and extremely high requirements for model accuracy make it very limited in practical applications. Slight changes in model parameters may lead to inaccurate results, especially in the face of multi-dimensional and dynamic environments, the applicability of analytical solutions is greatly reduced. On the other hand, although the numerical integration method can guarantee the calculation accuracy, it has a long calculation time and high resource consumption, and is not suitable for use in military decision-making scenarios with high real-time requirements. Especially in large-scale ballistic simulation, the numerical integration method cannot meet the needs of rapid response. Finally, although the method combining intelligent optimization algorithm and neural network can solve the problem of computational efficiency to a certain extent, they still have the problems of high algorithm complexity and large computational resource requirements. In addition, this type of method has a strong dependence on the initial parameters, and it is easy to fall into the local optimal solution during the search process, and cannot guarantee global optimality. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method, system, computer device and storage medium for rapid trajectory calculation, so as to solve the problem that traditional trajectory calculation methods cannot meet the real-time requirements of missile combat mission planning and decision-making, thereby improving the accuracy of our missile trajectory prediction while achieving the effect of rapid millisecond-level calculation of the trajectory.

[0006] In a first aspect, the present invention provides a method for fast trajectory calculation, the method comprising: generating trajectory calculation parameters based on rasterization, and calculating the trajectory calculation parameters to obtain a plurality of original trajectory data, performing feature extraction and segmentation processing on the original trajectory data to obtain a sample data set, the sample data set comprising a trajectory feature data set and a segmented trajectory scatter point data set;

[0007] The sample data set is used to train a preset trajectory solution model to obtain a trained trajectory solution model, wherein the trajectory solution model includes a front-segment trajectory sub-model, a middle-segment trajectory sub-model, a rear-segment trajectory sub-model, and a trajectory feature sub-model, wherein the front-segment trajectory sub-model, the middle-segment trajectory sub-model, and the rear-segment trajectory sub-model are trained using corresponding segmented trajectory scattered point data sets, and the trajectory feature sub-model is trained using the trajectory feature data set;

[0008] The parameters to be solved are input into the trained trajectory solution model for solution to obtain trajectory scatter data.

[0009] Furthermore, the steps of generating trajectory calculation parameters based on rasterization and solving the trajectory calculation parameters to obtain a plurality of original trajectory data include:

[0010] According to the longitude, latitude and altitude of the launch point and the target point, the range and geodetic azimuth are calculated by geodetic inverse calculation, and the latitude of the launch point, the altitude of the launch point, the range, the geodetic azimuth and the altitude of the target point are used as trajectory calculation parameters;

[0011] Determining a value range of the trajectory calculation parameters, and performing raster calculation of the trajectory calculation parameters at preset intervals to obtain multiple groups of trajectory calculation parameters;

[0012] Through the standard ballistic calculation program, multiple sets of ballistic calculation parameters are calculated to obtain multiple original ballistic data.

[0013] Furthermore, the step of extracting features and performing segmentation processing on the original trajectory data to obtain a sample data set includes:

[0014] Extracting features from the original trajectory data to obtain trajectory feature data, wherein the trajectory feature data includes a first-level shutdown time, a total flight time, and an aiming azimuth;

[0015] Obtaining a trajectory characteristic data set according to the trajectory calculation parameters and the trajectory characteristic data;

[0016] Expanding the original trajectory data in time sequence to obtain trajectory scatter data, and segmenting the trajectory scatter data using a segmentation strategy to obtain segmented trajectory scatter data;

[0017] A segmented trajectory scatter point data set is obtained according to the trajectory calculation parameters and the segmented trajectory scatter point data.

[0018] Furthermore, the step of segmenting the trajectory scattered point data by adopting a segmentation strategy to obtain segmented trajectory scattered point data comprises:

[0019] Performing a duration analysis on the flight phase of the missile flight trajectory according to the first-stage shutdown time and the total flight time, wherein the flight phase includes a front stage, a middle stage, and a rear stage;

[0020] Segmenting the trajectory scatter point data according to the duration of each flight phase to obtain segmented trajectory scatter point data;

[0021] Among them, the segmented trajectory scatter data includes front-segment scatter data, middle-segment scatter data and back-segment scatter data, the front-segment scatter data is the trajectory scatter data between zero time and the first shutdown time, the middle-segment scatter data is the trajectory scatter data between the first shutdown time and a preset middle time, and the back-segment scatter data is the trajectory scatter data between the middle time and the total flight time.

[0022] Furthermore, the step of inputting the parameters to be solved into the trained trajectory solution model for solution to obtain trajectory scattered point data includes:

[0023] Inputting the parameters to be solved into the trajectory characteristic sub-model to obtain the total flight time;

[0024] According to the preset segmentation time and the total flight time, the parameters to be solved are segmented to obtain the segmented parameters to be solved, and the segmentation time is determined by the time characteristics of the segmented trajectory scattered point data set;

[0025] The segmented parameters to be solved and the corresponding segmented moments are used as input parameters, and the input parameters are correspondingly input into the front segment trajectory sub-model, the middle segment trajectory sub-model and the back segment trajectory sub-model to obtain the front segment scattered point data, the middle segment scattered point data and the back segment scattered point data;

[0026] The front-end scattered point data, the middle-end scattered point data and the back-end scattered point data are spliced ​​to obtain the trajectory scattered point data of the parameter to be solved.

[0027] Furthermore, each sub-model of the trajectory solution model is constructed using a deep neural network model, and the loss function of each sub-model of the trajectory solution model adopts a mean square error loss function based on a weight matrix.

[0028] Furthermore, the loss function is expressed by the following formula:

[0029]

[0030] In the formula, m represents the number of data rows, sum represents the sum of each element of the matrix, L represents the mean square error loss matrix, and ω represents the weight matrix.

[0031] In a second aspect, the present invention provides a trajectory rapid solution system, the system comprising:

[0032] A sample set construction module, used to generate trajectory calculation parameters based on rasterization, and solve the trajectory calculation parameters to obtain multiple pieces of original trajectory data, perform feature extraction and segmentation processing on the original trajectory data to obtain a sample data set, wherein the sample data set includes a trajectory feature data set and a segmented trajectory scatter point data set;

[0033] A model building module, used to train a preset trajectory solution model using the sample data set to obtain a trained trajectory solution model, wherein the trajectory solution model includes a front-segment trajectory sub-model, a middle-segment trajectory sub-model, a rear-segment trajectory sub-model and a trajectory feature sub-model, wherein the front-segment trajectory sub-model, the middle-segment trajectory sub-model and the rear-segment trajectory sub-model are trained using corresponding segmented trajectory scattered point data sets, and the trajectory feature sub-model is trained using the trajectory feature data set;

[0034] The trajectory calculation module is used to input the parameters to be calculated into the trained trajectory calculation model for calculation to obtain trajectory scatter data.

[0035] In a third aspect, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0036] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0037] The present invention provides a method, system, computer device and storage medium for rapid trajectory calculation. The present invention constructs a trajectory calculation model based on a deep neural network model and adopts a segmented training strategy, which can effectively improve the learning efficiency of the model, the stability and accuracy of the prediction, and achieve millisecond-level rapid calculation capabilities while ensuring the prediction accuracy, which can meet the real-time requirements of combat decision-making and provide strong technical support for rapid decision-making in actual combat environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1is a schematic diagram of the process of the fast trajectory calculation method in an embodiment of the present invention;

[0039] Figure 2 is another schematic flow chart of the method for rapid trajectory calculation in an embodiment of the present invention;

[0040] Figure 3 is a schematic diagram of an Earth-centered Earth-fixed system and a launch coordinate system in an embodiment of the present invention;

[0041] Figure 4 is a gridded schematic diagram of the azimuth and range of a large area in an embodiment of the present invention;

[0042] Figure 5 is a schematic diagram of the generation process of a sample data set in an embodiment of the present invention;

[0043] Figure 6 is a schematic diagram of error results under different network structures in an embodiment of the present invention;

[0044] Figure 7 is a schematic diagram of a learning rate determination experiment in an embodiment of the present invention;

[0045] Figure 8 is a schematic diagram of an activation function image in an embodiment of the present invention;

[0046] Fig. 9 is a schematic diagram of learning rate variation in an embodiment of the present invention;

[0047] Fig.10 is a schematic diagram of the comparison results between the actual trajectory and the predicted trajectory in the comparative experiment of the embodiment of the present invention;

[0048] Fig.11 is a schematic diagram of a prediction error curve in a comparative experiment of an embodiment of the present invention;

[0049] Fig.12 is a schematic diagram of error statistics results corresponding to ballistic data in a comparative experiment of an embodiment of the present invention;

[0050] Fig.13 is a schematic structural diagram of a fast trajectory calculation system in an embodiment of the present invention;

[0051] Fig.14 It is a diagram of the internal structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] See also Figure 1 The first embodiment of the present invention provides a method for rapid trajectory calculation, which includes steps S10 to S30:

[0054] Step S10, generating trajectory calculation parameters based on rasterization, and solving the trajectory calculation parameters to obtain a plurality of original trajectory data, performing feature extraction and segmentation processing on the original trajectory data to obtain a sample data set, wherein the sample data set includes a trajectory feature data set and a segmented trajectory scatter point data set;

[0055] Step S20, using the sample data set to train a preset trajectory solution model to obtain a trained trajectory solution model, wherein the trajectory solution model includes a front-segment trajectory sub-model, a middle-segment trajectory sub-model, a rear-segment trajectory sub-model, and a trajectory feature sub-model, wherein the front-segment trajectory sub-model, the middle-segment trajectory sub-model, and the rear-segment trajectory sub-model are trained using corresponding segmented trajectory scattered point data sets, and the trajectory feature sub-model is trained using the trajectory feature data set;

[0056] Step S30, inputting the parameters to be solved into the trained trajectory solving model for solving, and obtaining trajectory scatter data.

[0057] The present invention provides a method for quickly solving the trajectory of our missiles. Currently, accurate trajectory data can be obtained through traditional various-element solving methods. The various-element solving methods are essentially numerical integration algorithms with a solving time of seconds, and as the trajectory range increases, the solving time will become slower. In large-scale trajectory calculation simulations, traditional various-element solving methods are simply unable to meet the needs of rapid combat response. In order to achieve millisecond-level rapid trajectory solving under the premise of meeting certain solving accuracy requirements, the present invention provides a fast-stage trajectory method based on a deep neural network, and its overall process is as follows: Figure 2As shown, first, the trajectory calculation parameters are determined according to the basic theory of ballistics, and the trajectory calculation parameters are rasterized at appropriate intervals within a certain range, and the arrangement and combination are used as the input of the traditional element solution method, thereby generating a large amount of trajectory data. The present invention adopts a segmented training strategy, that is, each trajectory data is divided into a front segment, a middle segment and a terminal segment, so it is necessary to pre-process the generated original trajectory data to form four sets of scattered point data of the front segment front, the middle segment mid and the rear segment end of the trajectory and the trajectory feature data, and respectively train the neural network to generate four neural network models: the front segment trajectory sub-model Model_front, the middle segment trajectory sub-model Model_mid, the rear segment trajectory sub-model Model_end and the trajectory feature sub-model Model_feature. When predicting, the prediction data of the four models are integrated in a certain logical order to form the final output. The learning-based method can achieve rapid generation of trajectory within millimeter time under the premise of ensuring prediction accuracy, and meet the real-time nature of combat decision-making. Based on the overall process, each step is described in detail below.

[0058] In a preferred embodiment, a rasterization-based trajectory data generation method is used to generate original trajectory data, and the specific steps include:

[0059] According to the longitude, latitude and altitude of the launch point and the target point, the range and geodetic azimuth are calculated by geodetic inverse calculation, and the latitude of the launch point, the altitude of the launch point, the range, the geodetic azimuth and the altitude of the target point are used as trajectory calculation parameters;

[0060] Determining a value range of the trajectory calculation parameters, and performing raster calculation of the trajectory calculation parameters at preset intervals to obtain multiple groups of trajectory calculation parameters;

[0061] Through the standard ballistic calculation program, multiple sets of ballistic calculation parameters are calculated to obtain multiple original ballistic data.

[0062] In this embodiment, it is first necessary to determine the required ballistic calculation parameters. The coordinate system is a reference datum selected to describe the position and motion law of the missile. The coordinate systems commonly used to describe the position of ballistic scatter points are the Earth-centered Earth-fixed system and the launch coordinate system. The sample data involved in the present invention are all generated in the launch coordinate system. Therefore, the launch coordinate system is used as the basic reference system to discuss how to determine the trajectory calculation parameters.

[0063] like Figure 3 As shown, X E Y E Z E The coordinate system is the Earth-centered Earth-fixed system, X F Y F Z FThe coordinate system is the launch coordinate system, and the missile launch point is O F , O F Y F The axis is the plumb line through the launch point, with upward being positive; F X F Axis and O F Y F The axis is vertical and points in the aiming direction. The angle A formed by it and the due north direction of the astronomical meridian plane of the launch point is T is called the azimuth angle of the transmitting point; F Z F Axis and O F X F ,O F Y F The axes form a right-hand rectangular coordinate system. Therefore, when the latitude and longitude (i.e. longitude, latitude and altitude) of the launch point and the target point are determined, the geodetic azimuth can be calculated by geodetic inverse calculation, and the launch coordinate system position can be completely determined, thereby determining a standard trajectory curve.

[0064] Since a large number of samples need to be generated, the trajectory calculation parameters can be rasterized at certain intervals and arranged and combined, thereby generating multiple groups of calculation parameters. In order to make the sample coverage wide and persuasive, the premise of rasterization needs to determine the range of each calculation parameter. If the longitude and latitude of the launch point and the target point are used as calculation parameters, the longitude and latitude range of the target point cannot be accurately determined, so the processing method of the present invention is to use the earth back calculation, calculate the range and the earth azimuth (the target point is relative to the launch point) through the longitude and latitude of the launch point and the target point, and use the longitude and latitude of the launch point, the range, the earth azimuth and the target point elevation as calculation parameters, so that all calculation parameters can easily determine the range. Further consideration, if the earth is regarded as a homogeneous rotating ellipsoid and the flight environment corresponding to each flight trajectory is the same, the longitude change of the launch point has little effect on the flight trajectory, that is to say, after other launch parameters are determined, the standard trajectory at each longitude is consistent. Therefore, the launch point latitude, the launch point elevation, the range, the earth azimuth and the target point elevation are finally determined as calculation parameters to generate samples. In order to facilitate the subsequent description, the symbols of various parameters involved in the present invention are defined here, as shown in Table 1.

[0065] Table 1 Related parameter symbol definition table

[0066] symbol definition lat_l Latitude of launch point lon_l Longitude of launch point alt_l Launch point height lat_t Target point latitude lon_t Target point longitude alt_t Target point height R Range A Great local angle <![CDATA[T0]]> Level 1 shutdown time <![CDATA[T1]]> Total flight time Az Aiming azimuth t Track time x,y,z Three-dimensional coordinates in the launch coordinate system

[0067] Since the method provided by the present invention uses a deep neural network for trajectory prediction, a large amount of sample data is required for training of the deep neural network. The quality of the sample data determines the upper limit of the model prediction performance, so the training samples should cover all possibilities of the problem as much as possible. Therefore, on the basis of determining the trajectory calculation parameters, the value range of each parameter is determined, and then the rasterized calculation parameters are calculated at reasonable intervals. According to the latitude range of the border and the missile range, the range of the five calculation parameters is determined and the appropriate interval is given. The interval here is determined after the comprehensive computing platform hardware conditions and the model training effects corresponding to different intervals, as shown in Table 2, where deg (degrees) represents the angle unit and km (kilometers) represents the length unit.

[0068] Table 2 Calculation parameter rasterization range and interval table

[0069]

[0070]

[0071] After rasterization according to the range and interval in Table 2, a total of 599,508 sets of trajectory calculation parameters will be generated. If all calculation parameters are valid, 599,508 trajectory data will be generated. After determining the launch point position, the rasterization diagram of the range and the azimuth angle is as follows: Figure 4 As shown, Figure 4 (a) is a schematic diagram of the azimuth gridding of a large area. Figure 4 (b) is a schematic diagram of range rasterization.

[0072] After rasterization to obtain multiple sets of trajectory calculation parameters, the latitude and longitude of the launch point and the target point are first obtained through geodetic calculation. The reason for the conversion is that the input interface of the standard trajectory calculation program is the latitude and longitude of the launch point and the target point. After the conversion, the trajectory is solved by the standard trajectory calculation program. The principle of the standard trajectory calculation program is to solve it based on the missile motion model by performing a fourth-order Runge-Kutta method integral operation. The specific principle can refer to the conventional solution steps, which will be described one by one here.

[0073] After obtaining multiple pieces of raw trajectory data through the above steps, further processing is required to generate a sample data set. The specific steps include:

[0074] Extracting features from the original trajectory data to obtain trajectory feature data, wherein the trajectory feature data includes a first-level shutdown time, a total flight time, and an aiming azimuth;

[0075] Obtaining a trajectory characteristic data set according to the trajectory calculation parameters and the trajectory characteristic data;

[0076] Expanding the original trajectory data in time sequence to obtain trajectory scatter data, and segmenting the trajectory scatter data using a segmentation strategy to obtain segmented trajectory scatter data;

[0077] A segmented trajectory scatter point data set is obtained according to the trajectory calculation parameters and the segmented trajectory scatter point data.

[0078] In this embodiment, the sample data set generation process is as follows: Figure 5 As shown in FIG. 1 , after the original trajectory data is obtained through the standard trajectory calculation program, the trajectory features are firstly extracted from the original trajectory data. In this embodiment, the trajectory feature data includes the first-level shutdown time, the total flight time and the aiming azimuth, that is, [T0 T1 Az]. Then, the previously determined trajectory calculation parameters are used as input parameters, and the trajectory feature data are used as label values ​​to form a trajectory feature data set, that is, [lat_l alt_l AR alt_t T0 T1 Az]. In other words, for the trajectory feature data sample, the input and output of the neural network are respectively: X input ={lat_l,alt_l,A,R,alt_t},X output ={T0,T1,Az}.

[0079] Then each original trajectory data is expanded in time series to obtain trajectory scatter data, which can be expressed as: [t i x i y i z i ], where t i Indicates the time of the i-th scattered data, x i ,y i and z i Represents the three-dimensional coordinates of the i-th scattered point data in the emission coordinate system. Then, a segmentation strategy is used to segment the trajectory scattered point data to obtain segmented trajectory scattered point data. The specific steps include:

[0080] Performing a duration analysis on the flight phase of the missile flight trajectory according to the first-stage shutdown time and the total flight time, wherein the flight phase includes a front stage, a middle stage, and a rear stage;

[0081] Segmenting the trajectory scatter point data according to the duration of each flight phase to obtain segmented trajectory scatter point data;

[0082] Among them, the segmented trajectory scatter data includes front-segment scatter data, middle-segment scatter data and back-segment scatter data, the front-segment scatter data is the trajectory scatter data between zero time and the first shutdown time, the middle-segment scatter data is the trajectory scatter data between the first shutdown time and a preset middle time, and the back-segment scatter data is the trajectory scatter data between the middle time and the total flight time.

[0083] In the present embodiment, the segmentation strategy is designed based on the segmentation principle of the missile trajectory. Specifically, according to whether the missile main engine works or not, the flight trajectory is divided into two sections: the active section and the passive section. The passive section can be divided into the free flight section and the reentry flight section according to the size of the aerodynamic force received by the aircraft. In general, the missile flight process is divided into the active section, the free section, and the reentry section in turn, and the free section flight time accounts for more than 80% to 90% of the whole trajectory. The purpose of segmenting the flight trajectory is to use different methods to solve the differential equation of motion in different flight sections to obtain the objective law of the aircraft motion. As for model training, segmentation training can better fit the motion law of different flight sections, thereby obtaining a model with higher overall prediction accuracy, which is the theoretical support for adopting the segmentation strategy in the present embodiment.

[0084] In this embodiment, for the trajectory scatter data, the segmentation processing rule is: each trajectory scatter data from time 0 to the first shutdown time is used as the front-end scatter data, and the corresponding active segment of the missile flight trajectory. Considering that the free segment usually accounts for more than 80% to 90% of the total trajectory time, the maximum flight time in the trajectory sample is counted at the same time, and the total duration of the active segment and the reentry segment can be obtained according to the maximum flight time and the time proportion of the free segment. Because the duration of the active segment can be determined based on the first shutdown time, the duration of the reentry segment can be obtained. Assuming that the maximum flight time counted is 666.137s, the total time of the reentry segment and the active segment is at most about 133s. Taking the first-level shutdown time as the end time of the active segment, the minimum first-level shutdown time in all samples is determined to be 58.6s, so the maximum duration of the reentry segment in the sample is about 74.4s. The time interval between trajectory scatter data is 1 second, so the last 80 rows of scatter data of each trajectory are selected as the rear-segment scatter data, which corresponds to the reentry segment of the missile flight trajectory, and the remaining intermediate scatter data are selected as the mid-segment scatter data, which corresponds to the free segment of the missile flight trajectory.

[0085] After segmenting the trajectory scatter data, the trajectory calculation parameters and trajectory time are used as input parameters, and the trajectory scatter data is used as the label value. The segmented trajectory scatter data set is formed according to the segmentation results based on time t, that is, [lat_l alt_l AR alt_t txyz]. In other words, for the segmented sample data of the trajectory scatter data set, the input and output of the neural network are consistent, which are: X input ={lat_l,alt_l,A,R,alt_t,t},X output ={x,y,z}.

[0086] After the sample feature values ​​and label values ​​are determined, due to the inconsistency of the dimensions and value ranges between the features, if normalization is not performed, the contour map of the loss function will be elliptical, which will affect the direction of gradient descent, slow down convergence or fail to converge. In order to eliminate the influence of feature dimensions and numerical scales, the maximum and minimum normalization method is used to linearly transform the original data by column so that all feature data fall in the [0,1] interval. The normalization formula is as follows:

[0087]

[0088] Among them, X is the original data, X norm is the normalized data, X min is the minimum value of each feature, X max is the maximum value of each feature.

[0089] In this embodiment, the trajectory solution model consists of four sub-models, namely, the front-end trajectory sub-model, the middle-end trajectory sub-model, the back-end trajectory sub-model and the trajectory feature sub-model. Each sub-model is constructed using a deep neural network DNN, and its related network parameters are shown in Table 3.

[0090] Table 3 Network model related parameters

[0091]

[0092]

[0093] Based on the above model parameters, the optimal parameters of the model are determined through training. Specifically, the network structure is an important factor affecting the training effect of the neural network, and it lays the lower limit of the training effect of the neural network. Changing the number of hidden layers or the number of neural units in each layer is essentially the same, which increases the number of neurons in the overall network, thereby improving the complexity and fitting ability of the model. Too few hidden layers or neurons will lead to underfitting, while too many hidden layers or neurons may increase the training time and may lead to overfitting. Assuming that the number of neurons in each layer is the same, the present invention determines the number of hidden layers and the number of units in each layer through cross-validation. It should be noted that the model parameters given in this embodiment are only preferred and not specifically limited, and will not be repeated later.

[0094] In this embodiment, the errors under different network structures are as follows: Figure 6 As shown, the training error corresponding to different combinations of layers and units is Figure 6 (a) and the test error is Figure 6 (b), the following conclusions can be drawn through comparison: First, with the increase of the number of layers, both the training error and the test error increase. Second, for networks with different numbers of layers, with the increase of the number of units, the training error and the test error show a trend of first decreasing and then increasing or first decreasing and then stabilizing. Based on the above analysis, the network structure is determined to be a network structure with 2 hidden layers and 256 units per layer. Under this structure, the training error and the test error are very small, indicating that the model learning effect and generalization are relatively good.

[0095] The learning rate (LR) is a parameter that affects the convergence speed and learning effect of deep neural networks. In order to find a suitable LR so that the model can obtain the best loss value without reducing the training speed, this embodiment uses different LRs to experiment with the network model. Figure 7 (a) and Figure 7 (b) represents the training error and test error under different learning rates, respectively. Figure 7 The following conclusions can be drawn from the results: a smaller LR can reduce the fluctuation of training and validation losses, but if LR is too small, the model convergence speed will become very slow. Therefore, in order to strike a balance between accuracy and computational burden, this embodiment selects LR=0.001 for subsequent experiments.

[0096] Regarding the setting of the loss function, since fast trajectory prediction is essentially a regression problem, the mean square error function is usually used as the loss function. However, for the trajectory scattered data [x, y, z] in the launch coordinate system, the z label data value is relatively small compared to x and y. In the error back propagation process, the traditional mean square error function will reduce the influence of the error value corresponding to z on the weight information update. Therefore, a weight matrix ω is needed to increase the error weight corresponding to the z label value, so that the z label value can be better learned. The specific setting of the loss function is explained below:

[0097] Define the number of data rows in a batch as m, then the corresponding prediction value is:

[0098]

[0099] The true value is:

[0100]

[0101] The corresponding variance loss matrix is:

[0102]

[0103] Let the weight matrix ω = [ω1ω2ω3], where ω1, ω2, ω3 are the error weights corresponding to x, y, and z respectively, then the final average loss function is:

[0104]

[0105] Among them, sum() means summing up all elements of the matrix, ω = [0.3, 0.3, 0.4].

[0106] In order to avoid the problem of neuron death, the LeakyReLU function is used as the activation function in this embodiment, and its image is as follows: Figure 8 As shown, its specific expression is:

[0107]

[0108] Here, x represents the input parameter.

[0109] Compared with other optimizers, such as stochastic gradient descent, the Adam optimizer performs well when the data set is large. Since the data sample capacity involved in the present invention is large, the Adam optimizer is used. In order to better improve the training efficiency and enhance the training effect, a dynamically changing learning rate is adopted in this embodiment, that is, in the early stage of training, the learning rate increases linearly, and after increasing to a specified value, the learning rate decays cosine-like. The specific learning rate change trend is as follows: Fig. 9As shown, the vertical axis represents the learning rate, and the maximum value is set to 0.05; the horizontal axis represents the epoch, i.e. the number of rounds, and the total number of epochs is 200. In the first 1 / 10 of the epochs, the learning rate increases linearly from 0 to 0.05, and then decays from 0.05 cosine to 0.

[0110] In this embodiment, the four sub-models Model of the trajectory solution model, namely the front trajectory sub-model Model_front, the middle trajectory sub-model Model_mid, the rear trajectory sub-model Model_end and the trajectory feature sub-model Model_feature, are all trained according to the above-mentioned parameter adjustment ideas, and finally the optimal model structure and parameters are determined as shown in Table 4 below.

[0111] Table 4 Model structure and training parameters

[0112]

[0113] After completing all the training, the final ballistic solution model includes four sub-models: Model_Front, Model_Mid, Model_End and Model_Feature. For Model_Feature, which is the trajectory feature sub-model, the data used is the total flight time T1 of the missile output by the sub-model. For the segmented trajectory sub-model, in this embodiment, the ballistic time range of each segment is determined by the maximum and minimum values ​​of the time feature in the corresponding training set. Since the first-level shutdown time of each trajectory is not the same, the first-level shutdown time is only used as a preliminary division of the original data set. In this embodiment, the trajectory time in the sample set is analyzed to determine the time values ​​of different segments of the input parameters when making predictions. The segmented trajectory scatter data is represented by Front samples, Mid samples and End samples. It is assumed that the value range of the time feature in the Front sample set is [0,t front_max ], the value range of the time feature in the Mid sample set is [t mid_min ,t mid_max ], the value range of the time feature in the End sample set is [t end_min ,t end_max ], then during inference prediction, the time ranges of the three model inputs, Model_Front, Model_Mid, and Model_End, are determined as follows:

[0114]

[0115] Therefore, the input of the Model_Front model is expressed as:

[0116]

[0117] The input of the Model_Mid model is expressed as:

[0118]

[0119] The input of the Model_End model is expressed as:

[0120]

[0121] For the trained trajectory calculation model, the calculation steps for the parameters to be calculated include:

[0122] Inputting the parameters to be solved into the trajectory characteristic sub-model to obtain the total flight time;

[0123] According to the preset segmentation time and the total flight time, the parameters to be solved are segmented to obtain the segmented parameters to be solved, and the segmentation time is determined by the time characteristics of the segmented trajectory scattered point data set;

[0124] The segmented parameters to be solved and the corresponding segmented moments are used as input parameters, and the input parameters are correspondingly input into the front segment trajectory sub-model, the middle segment trajectory sub-model and the back segment trajectory sub-model to obtain the front segment scattered point data, the middle segment scattered point data and the back segment scattered point data;

[0125] The front-end scattered point data, the middle-end scattered point data and the back-end scattered point data are spliced ​​to obtain the trajectory scattered point data of the parameter to be solved.

[0126] In this embodiment, based on the trajectory calculation model established by the above-mentioned deep neural network DNN, a complete trajectory prediction process can be expressed as:

[0127] ① Load four models: Model_Front, Model_Mid, Model_End and Model_Feature

[0128] ② Initialization input: X = {lat_l, alt_l, A, R, alt_t};

[0129] Normalize X: X→X normalized ;

[0130] ③ Through Model_Feature(X normalized ) obtain the total flight time T1 of the missile;

[0131] ④ Initialize 3 sets of 2D arrays and name them as: X front , X mid , X end ;

[0132] According to the above three models input time range, the trajectory time t is divided, and according to the segmented time t, the input parameter X is divided into segmented input parameters, which are input into the corresponding two-dimensional arrays respectively to obtain three two-dimensional arrays X front , X mid , X end ;

[0133] Normalize the resulting two-dimensional array: X front →X1, X mid →X2, X end →X3;

[0134] ⑤ Obtain Y1, Y2 and Y3 in turn through Model_Front(X1), Model_Mid(X2) and Model_End(X3), and concatenate Y1, Y2 and Y3 by column to form complete ballistic scatter data.

[0135] In order to further verify the prediction accuracy of the trajectory calculation method provided by the present invention, the present invention will conduct a comparative experiment to compare the actual trajectory calculated by traditional parameters with the predicted trajectory obtained by DNN prediction provided by the present invention, and analyze the prediction accuracy of the calculation method provided by the present invention based on the comparison results. Assuming that the randomly given parameters to be solved are [42°, 1000m, 210°, 611117m, 4000m], the comparison results of the actual trajectory and the predicted trajectory are as follows: Fig.10 As shown, Fig.10 (a) shows the distance comparison in the X, Y, and Z directions in the launch coordinate system, and the distance in the Z direction is basically 0; Fig.10 (b) represents the prediction error in the X, Y, and Z directions in the launch coordinate system, based on Fig.10 (a) and Fig.10 (b) It can be seen that since the values ​​in the Z direction are too different from those in the X and Y directions, if the normal mean square error is calculated in actual training, the error obtained in the XY direction is too large, and the convergence effect of the Z direction learning may be poor. In view of this problem, the present invention designs an error weight matrix, and the existing error results in the Z direction can show the effectiveness of the error weight. The errors in the three directions of XYZ show an increasing trend with the flight time, because with the flight time, the distance in the X direction is constantly increasing. Therefore, when learning the data in the late flight, the corresponding mean square error is larger in magnitude, so the convergence effect is worse than that in the early flight, so the final error will increase slightly with the flight time.

[0136] Fig.10 (c) shows the comparison between the actual trajectory and the predicted trajectory of the missile in the launch coordinate system. Fig.10(d) represents the Euclidean distance error corresponding to the actual trajectory, which can be regarded as the prediction error in three-dimensional space. It can be seen that the Euclidean distance also increases slightly with the flight time. The reason for the increase is Fig.10 (b) The reason for the slight increase in the error in the XYZ direction is the same. In general, the Euclidean distance corresponding to most moments is below 100, and the Euclidean distance in the later stage of flight is also between 100 and 200. This prediction accuracy is feasible for providing trajectory simulation verification for related intelligent decision-making systems.

[0137] In order to further verify the generalization ability of the model generated by the present invention, Monte Carlo simulation and error analysis were performed in the comparative experiment. 1320 (5 latitude*2 height*2 height*11*6) groups of launch calculation parameters were randomly generated, and the trajectory data predicted by the model were compared with the trajectory data calculated by the standard calculation program to obtain the statistics of the relevant errors. Among them, T1 predicted by Model_Feature, Fig.11 This is the T1 prediction error curve for 1320 test cases. The maximum error does not exceed 0.33s, which meets the accuracy requirements for the overall trajectory prediction.

[0138] Next, we will count the maximum value max, minimum value min, median and average value mean of the absolute value of the error corresponding to each trajectory, and the resulting box plot is as follows: Fig.12 As shown, according to Fig.12 The box plot corresponding to the maximum value shows that the maximum error of each trajectory in the XYZ direction is about 200, which shows that the model prediction is stable and no abnormal prediction results appear. Fig.12 In the box plot corresponding to the median, the median of the XY direction error is mostly around 20, indicating that the prediction accuracy of most time series data is very high, which can also be seen from the box plot of the average value. The maximum, minimum, median and average value of the overall error of the Monte Carlo simulation under the segmentation strategy are given below, as shown in Table 5.

[0139] Table 5 Monte Carlo simulation error table under segmentation strategy

[0140] error Maximum Minimum median average value X direction 198.7625 0.0 20.2602 25.4981 Y direction 217.7961 0.0 21.3373 29.0140 Z direction 8.6060 0.0 1.2806 1.1997

[0141] The present invention uses a segmented training strategy to improve prediction accuracy. In order to illustrate the feasibility and effectiveness of the segmented strategy, based on the Monte Carlo simulation data, the model trained under the unsegmented strategy is used for prediction to obtain the relevant error statistics corresponding to the prediction results. The network structure and training parameters used for the unsegmented training are shown in Table 6 below.

[0142] Table 6 Network structure and training parameters used in training under the unsegmented strategy

[0143] Model num_layers num_units batch_size epochs lr ω n Unsegmented Model 4 256 1024 200 0.001 [0.3,0.3,0.4] 2

[0144] The specific prediction errors are shown in Table 7 below:

[0145] Table 7 Monte Carlo simulation error table under unsegmented strategy

[0146] error Maximum Minimum median average value X direction 1142.8187 0.0001 293.3989 361.3222 Y direction 2525.8000 0.0 393.5313 667.3912 Z direction 30.6585 0.0 0.3644 0.6067

[0147] According to the comparison between Table 7 and Table 5, it can be seen that the maximum and average values ​​of the model prediction error obtained without segmented training are about 10 times the model prediction error under the segmented training strategy. This shows that the segmented strategy is effective in reducing the overall prediction error and improving the prediction stability.

[0148] The goal of the present invention is to achieve millisecond-level rapid solution of the trajectory while ensuring accuracy, so as to meet the requirements of real-time combat mission planning and decision-making. In order to verify the rapidity of the model solution, the solution time corresponding to 1 trajectory and 100 trajectories is statistically analyzed respectively. In order to avoid random errors, when statistically predicting the time, the test program is run 10 times in sequence, and the time results obtained from the 10 runs are averaged as the final result. At the same time, the solution time corresponding to the standard trajectory solution program is statistically compared, and the specific results are shown in Table 8:

[0149] Table 8 Ballistic solution time statistics

[0150] method 1 trajectory 100 ballistics Standard ballistic solver 1.585 162.257 The method provided by the present invention 0.033 2.965

[0151] It can be seen from Table 8 that the DNN-based trajectory solving method provided by the present invention can achieve millisecond-level trajectory solving, and from the solving time of 100 trajectories, it can be seen that the DNN-based trajectory solving method has great advantages in large-scale trajectory solving.

[0152] This embodiment provides a method for rapid trajectory calculation. The present invention combines deep learning technology with the actual trajectory segmentation characteristics, and proposes a rapid trajectory prediction method based on a segmentation training strategy. Compared with the traditional trajectory calculation method, which has the problems of slow calculation speed, low prediction accuracy and poor adaptability, the present invention can achieve millisecond-level rapid trajectory calculation while ensuring prediction accuracy, meeting the real-time requirements of missile combat mission planning and decision-making, and has significant flexibility and practicality.

[0153] Specifically, the present invention adopts a segmented training strategy to independently model different stages of the trajectory. This method can conduct specialized training for the physical characteristics and nonlinear features of each stage, thereby improving the learning efficiency and prediction accuracy of the model. This strategy effectively solves the problems of difficult training and inaccurate prediction caused by model complexity in the prior art.

[0154] The present invention is based on the fusion design of multi-scale deep neural network models and implements the fusion of four interrelated deep neural network models to achieve accurate prediction of ballistic trajectories. The architecture and parameter design of these models have been carefully optimized so that they can adapt to changes in different input features and effectively improve the stability and accuracy of the overall prediction. And in the training process, in order to solve the impact of errors in different dimensions on model learning, the present invention introduces an error weight matrix. This design effectively balances the errors in each dimension, improves the learning effect and convergence of the model in each dimension, thereby further improving the prediction accuracy. While ensuring the prediction accuracy, the present invention achieves a fast solution capability at the millisecond level, meets the real-time requirements of combat decision-making, and provides strong technical support for rapid decision-making in actual combat environments.

[0155] See also Fig.13 Based on the same inventive concept, a fast trajectory calculation system proposed in the second embodiment of the present invention includes:

[0156] The sample set construction module 10 is used to generate trajectory calculation parameters based on rasterization, and solve the trajectory calculation parameters to obtain multiple original trajectory data, perform feature extraction and segmentation processing on the original trajectory data to obtain a sample data set, wherein the sample data set includes a trajectory feature data set and a segmented trajectory scatter point data set;

[0157] A model building module 20 is used to train a preset trajectory solution model using the sample data set to obtain a trained trajectory solution model, wherein the trajectory solution model includes a front-segment trajectory sub-model, a middle-segment trajectory sub-model, a rear-segment trajectory sub-model, and a trajectory feature sub-model, wherein the front-segment trajectory sub-model, the middle-segment trajectory sub-model, and the rear-segment trajectory sub-model are trained using corresponding segmented trajectory scattered point data sets, and the trajectory feature sub-model is trained using the trajectory feature data set;

[0158] The trajectory calculation module 30 is used to input the parameters to be calculated into the trained trajectory calculation model for calculation to obtain trajectory scatter data.

[0159] The technical features and technical effects of the trajectory rapid solution system proposed in the embodiment of the present invention are the same as those of the method proposed in the embodiment of the present invention, and are not described in detail here. Each module in the above-mentioned trajectory rapid solution system can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0160] In addition, an embodiment of the present invention further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0161] See also Fig.14 , an internal structure diagram of a computer device in an embodiment, the computer device can specifically be a terminal or a server. The computer device includes a processor, a memory, a network interface, a display and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a fast trajectory solution method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0162] It can be understood by those skilled in the art that Fig.14 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computing device may include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0163] In addition, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0164] In summary, an embodiment of the present invention proposes a method, system, computer device and storage medium for rapid trajectory solution. The method generates trajectory calculation parameters based on rasterization, solves the trajectory calculation parameters, obtains multiple original trajectory data, extracts features and performs segmentation processing on the original trajectory data, and obtains a sample data set, wherein the sample data set includes a trajectory feature data set and a segmented trajectory scatter data set; uses the sample data set to train a preset trajectory solution model to obtain a trained trajectory solution model, wherein the trajectory solution model includes a front-segment trajectory sub-model, a middle-segment trajectory sub-model, a rear-segment trajectory sub-model and a trajectory feature sub-model, wherein the front-segment trajectory sub-model, the middle-segment trajectory sub-model and the rear-segment trajectory sub-model are trained using the corresponding segmented trajectory scatter data sets, and the trajectory feature sub-model is trained using the trajectory feature data set; the parameters to be solved are input into the trained trajectory solution model for solution to obtain trajectory scatter data. The present invention combines deep learning technology with the actual segmentation characteristics of trajectory, and proposes a rapid trajectory prediction method based on segmentation training strategy. It can achieve millisecond-level rapid trajectory solution while ensuring prediction accuracy, meet the real-time requirements of missile combat mission planning and decision-making, and has significant flexibility and practicality.

[0165] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0166] The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims

1. A fast trajectory calculation method, characterized in that: include: Generate trajectory calculation parameters based on rasterization, solve the trajectory calculation parameters to obtain multiple pieces of original trajectory data, perform feature extraction and segmentation processing on the original trajectory data to obtain a sample data set, the sample data set includes a trajectory feature data set and a segmented trajectory scatter point data set; The sample data set is used to train a preset trajectory solution model to obtain a trained trajectory solution model, wherein the trajectory solution model includes a front-segment trajectory sub-model, a middle-segment trajectory sub-model, a rear-segment trajectory sub-model, and a trajectory feature sub-model, wherein the front-segment trajectory sub-model, the middle-segment trajectory sub-model, and the rear-segment trajectory sub-model are trained using corresponding segmented trajectory scattered point data sets, and the trajectory feature sub-model is trained using the trajectory feature data set; The parameters to be solved are input into the trained trajectory solution model for solution to obtain trajectory scatter data.

2. The fast trajectory calculation method according to claim 1, characterized in that: The steps of generating trajectory calculation parameters based on rasterization and solving the trajectory calculation parameters to obtain a plurality of original trajectory data include: According to the longitude, latitude and altitude of the launch point and the target point, the range and geodetic azimuth are calculated by geodetic inverse calculation, and the latitude of the launch point, the altitude of the launch point, the range, the geodetic azimuth and the altitude of the target point are used as trajectory calculation parameters; Determining a value range of the trajectory calculation parameters, and performing raster calculation of the trajectory calculation parameters at preset intervals to obtain multiple groups of trajectory calculation parameters; Through the standard ballistic calculation program, multiple sets of ballistic calculation parameters are calculated to obtain multiple original ballistic data.

3. The fast trajectory calculation method according to claim 1, characterized in that: The step of extracting features and performing segmentation processing on the original trajectory data to obtain a sample data set comprises: Extracting features from the original trajectory data to obtain trajectory feature data, wherein the trajectory feature data includes a first-level shutdown time, a total flight time, and an aiming azimuth; Obtaining a trajectory characteristic data set according to the trajectory calculation parameters and the trajectory characteristic data; Expanding the original trajectory data in time sequence to obtain trajectory scatter data, and segmenting the trajectory scatter data using a segmentation strategy to obtain segmented trajectory scatter data; A segmented trajectory scatter point data set is obtained according to the trajectory calculation parameters and the segmented trajectory scatter point data.

4. The fast trajectory calculation method according to claim 3, characterized in that: The step of segmenting the trajectory scattered point data by adopting a segmentation strategy to obtain the segmented trajectory scattered point data comprises: Performing a duration analysis on the flight phase of the missile flight trajectory according to the first-stage shutdown time and the total flight time, wherein the flight phase includes a front segment, a middle segment, and a rear segment; Segmenting the trajectory scatter point data according to the duration of each flight phase to obtain segmented trajectory scatter point data; Among them, the segmented trajectory scatter data includes front-segment scatter data, middle-segment scatter data and back-segment scatter data, the front-segment scatter data is the trajectory scatter data between zero time and the first shutdown time, the middle-segment scatter data is the trajectory scatter data between the first shutdown time and a preset middle time, and the back-segment scatter data is the trajectory scatter data between the middle time and the total flight time.

5. The fast trajectory calculation method according to claim 3, characterized in that: The step of inputting the parameters to be solved into the trained trajectory solving model for solving to obtain trajectory scattered point data comprises: Inputting the parameters to be solved into the trajectory characteristic sub-model to obtain the total flight time; According to the preset segmentation time and the total flight time, the parameters to be solved are segmented to obtain the segmented parameters to be solved, and the segmentation time is determined by the time characteristics of the segmented trajectory scattered point data set; The segmented parameters to be solved and the corresponding segmented moments are used as input parameters, and the input parameters are correspondingly input into the front segment trajectory sub-model, the middle segment trajectory sub-model and the back segment trajectory sub-model to obtain the front segment scattered point data, the middle segment scattered point data and the back segment scattered point data; The front-end scattered point data, the middle-end scattered point data and the back-end scattered point data are spliced ​​to obtain the trajectory scattered point data of the parameter to be solved.

6. The fast trajectory calculation method according to claim 1, characterized in that: Each sub-model of the trajectory solution model is constructed using a deep neural network model, and the loss function of each sub-model of the trajectory solution model adopts a mean square error loss function based on a weight matrix.

7. The fast trajectory calculation method according to claim 6, characterized in that: The loss function is expressed by the following formula: In the formula, m represents the number of data rows, sum represents the sum of each element of the matrix, L represents the mean square error loss matrix, and ω represents the weight matrix.

8. A fast trajectory calculation system, characterized in that: include: A sample set construction module, used to generate trajectory calculation parameters based on rasterization, and solve the trajectory calculation parameters to obtain multiple pieces of original trajectory data, perform feature extraction and segmentation processing on the original trajectory data to obtain a sample data set, wherein the sample data set includes a trajectory feature data set and a segmented trajectory scatter point data set; A model building module, used to train a preset trajectory solution model using the sample data set to obtain a trained trajectory solution model, wherein the trajectory solution model includes a front-segment trajectory sub-model, a middle-segment trajectory sub-model, a rear-segment trajectory sub-model and a trajectory feature sub-model, wherein the front-segment trajectory sub-model, the middle-segment trajectory sub-model and the rear-segment trajectory sub-model are trained using corresponding segmented trajectory scattered point data sets, and the trajectory feature sub-model is trained using the trajectory feature data set; The trajectory calculation module is used to input the parameters to be calculated into the trained trajectory calculation model for calculation to obtain trajectory scatter data.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images