A Cooperative Planning Method for Simultaneous Hit of Multiple Projectiles in an Electromagnetic Railgun

By using an MDN-MLP concatenated network and a direction error feedback algorithm, combined with the physical constraints of the electromagnetic railgun, rapid and accurate collaborative planning of multiple projectiles from the electromagnetic railgun was achieved. This solved the multi-mode ballistic inverse problem of uncontrolled projectiles from the electromagnetic railgun, meeting the real-time strike requirements of modern warfare.

CN123114017APending Publication Date: 2026-11-17DALIAN UNIV OF TECH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611621415.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies are unable to quickly and accurately solve the multi-mode ballistic inversion problem of uncontrolled projectiles for electromagnetic railguns, and cannot coordinate online planning for multiple projectiles to hit the same target simultaneously, thus failing to meet the strike requirements of new threats such as hypersonic cruise missiles and drone swarms.

Method used

A geometric correction algorithm combining MDN-MLP cascaded network and directional error feedback is adopted. The initial trajectory inclination angle and flight time are predicted through offline training. A greedy rule to maximize the number of cooperative strike sequences is designed to meet the physical constraints and real-time requirements of the electromagnetic gun.

Benefits of technology

This method enables multiple electromagnetic railgun projectiles to hit the same target simultaneously and accurately within milliseconds, meeting the requirements for high precision and real-time performance. It solves the problems of large computational load and large prediction error in traditional methods and is suitable for collaborative planning of uncontrolled projectiles.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114017A_ABST
    Figure CN123114017A_ABST
Patent Text Reader

Abstract

This invention provides a collaborative planning method for simultaneous hits of multiple projectiles from an electromagnetic railgun, belonging to the field of fire control and ballistic planning technology. This invention rapidly predicts the initial trajectory inclination angle and projectile flight time using an offline-trained MDN-MLP concatenated network, improves impact accuracy using a geometric correction algorithm based on direction error feedback, and generates collaborative strike sequences that satisfy firing interval constraints with the goal of maximizing the number of collaborative strike sequences. This method enables rapid solution of collaborative planning for multiple electromagnetic railgun projectiles simultaneously hitting the same target in modern warfare environments facing new threats such as hypersonic cruise missiles and UAV swarms. This invention is the first to achieve collaborative planning for simultaneous hits of multiple uncontrolled electromagnetic railgun projectiles, meeting the real-time requirements of rapid strike missions while ensuring engineering accuracy, and providing a candidate technical solution for electromagnetic railguns to intercept high-speed maneuvering targets.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of fire control and ballistic planning technology, and relates to a method for coordinated planning of multiple simultaneous hits of electromagnetic railgun projectiles, which can realize coordinated planning of multiple projectiles hitting the same target at the same time. Background Technology

[0002] At present, emerging threats such as hypersonic cruise missiles and drone swarms pose a severe challenge to air defense interception. Electromagnetic railguns have advantages such as high initial velocity, long range, and flexible response, but their unguided projectiles lack in-flight correction capabilities, resulting in a limited single-shot hit probability. Therefore, multiple projectiles need to hit the same target simultaneously to form an instantaneous barrage to improve strike effectiveness.

[0003] Achieving this goal faces two major challenges. First, solving for the initial ballistic parameters is a nonlinear inverse problem. The same target point corresponds to two types of trajectories: high-angle and low-angle. Traditional numerical traversal methods are computationally intensive and have poor real-time performance, and conventional neural networks are difficult to fit this multimodal distribution. Second, uncontrolled projectiles cannot be coordinated online, and parameter planning must be completed before launch. Furthermore, artillery has physical constraints such as firing intervals. For solving the inverse ballistic problem, existing technologies such as the parameter space discretization traversal method, although theoretically convergent, have computational costs that increase exponentially with resolution, making it difficult to meet the real-time requirements of rapid strikes. While deep learning-based prediction methods (Chinese invention patent CN119647286B "Method for Estimating Vehicle Dynamics Parameters Based on Machine Learning") have achieved some success in forward dynamics problems, they cannot handle the multimodal output characteristics of the inverse ballistic problem, and the prediction error is large in sparse areas of network training data or at ballistic boundaries, making it difficult to directly meet the high-precision requirements of coordinated strikes. To address the issue of multi-missile coordination, existing technologies mostly employ time-constrained terminal guidance laws or distributed cooperative guidance laws (Chinese Invention Patent CN116880560B "A Multi-Aircraft Time-Division Strike Cooperative Guidance Method"). These methods assume that the projectile has full-range controllability and inter-projectile information interaction, and are suitable for missiles or guided projectiles, but cannot be directly applied to completely uncontrolled electromagnetic projectiles.

[0004] Therefore, developing a collaborative planning method for uncontrolled projectiles that can quickly solve the multimodal distribution of the inverse ballistic problem, balance accuracy and real-time performance, and adapt to artillery constraints is of great significance for improving the strike effectiveness of electromagnetic railguns. Summary of the Invention

[0005] To address the technical challenges of low efficiency in solving existing ballistic inverse problems, difficulty in fitting multimodal distributions, and the inability of uncontrolled projectiles to coordinate online, this invention proposes a collaborative planning method for multiple simultaneous hits by electromagnetic railgun projectiles. This invention rapidly predicts the initial trajectory inclination angle and projectile flight time using an offline-trained MDN-MLP concatenated network, improves impact point accuracy using a geometric correction algorithm based on direction error feedback, and generates collaborative attack sequences that satisfy firing interval constraints with the goal of maximizing the number of collaborative attack sequences. This method enables rapid solution of collaborative planning for multiple electromagnetic railgun projectiles simultaneously hitting the same target in modern warfare environments facing new threats such as hypersonic cruise missiles and UAV swarms.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for coordinated planning of simultaneous hits by multiple projectiles in an electromagnetic railgun includes the following steps: Step 1: The solution process for the electromagnetic railgun multi-projectile coordinated attack planning problem is divided into two sub-problems: accurately solving for the initial parameters of a single-shot trajectory and generating a multi-projectile coordinated attack sequence; specifically: The initial parameters of a single-shot ballistic trajectory include the initial trajectory inclination angle of the projectile and the flight time of the projectile to reach the target point. The solution process includes multi-modal parameter prediction of the inverse ballistic problem and geometric correction of residual error. Step 2: Set the physical constraints that must be met when solving the initial ballistic parameters to ensure the feasibility of the solution in actual engineering.

[0007] When solving for initial trajectory parameters to plan coordinated attacks or projectile launches, electromagnetic railguns must satisfy the physical constraints proposed in this invention, including initial velocity discrete step constraints, initial trajectory inclination angle boundary constraints, trajectory rationality constraints, and impact point accuracy constraints. Specifically: The initial velocity discrete range constraint: The initial velocity of the electromagnetic railgun is determined by hardware conditions such as power supply energy and track length, and cannot be continuously and steplessly adjusted. In practical engineering, only a limited number of discrete ranges can be selected. Therefore, during parameter planning, the initial velocity value must be constrained to belong to the candidate discrete initial velocity range set. Let the candidate discrete initial velocity range set be... for: (1) in, This indicates the number of gears, which is determined by the actual performance of the electromagnetic gun. For the first The initial velocity value corresponding to each gear; therefore This is the initial velocity value corresponding to the first gear; This is the initial velocity value corresponding to the second gear; For the first The initial velocity value corresponding to each gear.

[0008] The initial trajectory inclination angle boundary constraint: the initial trajectory inclination angle of the projectile's outer trajectory. The elevation angle limit of the electromagnetic railgun launching mechanism needs to be met, and its range is set as follows: (2) in, The initial trajectory angle of the projectile; and These are the minimum and maximum pitch angles allowed by the electromagnetic railgun launching mechanism, respectively.

[0009] The ballistic rationality constraint is as follows: Due to the dimensionality reduction simplification achieved by employing the small angle-of-attack assumption in the external ballistic dynamics model, excessive angle-of-attack or its rate of change during flight will lead to significant lift nonlinearity and drastic changes in the moment coefficient, causing the model to fail and the solution results to lose physical meaning. Therefore, an upper limit constraint needs to be imposed on the projectile's angle of attack and its rate of change. (3) in, For the angle of attack of the projectile; Let be the rate of change of the projectile's angle of attack; and These are the set upper limit for angle of attack and the upper limit for rate of change of angle of attack, respectively.

[0010] The impact point accuracy constraint is as follows: to ensure the effectiveness of the strike, the Euclidean distance between the impact point and the target point corresponding to the calculated initial ballistic parameters must be less than a preset threshold. ,Right now: (4) in, The landing point corresponding to the calculated initial ballistic parameters; For the target point; The preset threshold value varies depending on the type of propellant used in the projectile. .

[0011] Step 3: After clarifying the physical constraints in Step 2, in order to obtain accurate projectile flight trajectory and time information, an external ballistic dynamics model is established based on the flight motion characteristics of the uncontrolled projectile, and a training dataset for training the MDN-MLP concatenated network is constructed accordingly; specifically: Step 3-1: Based on the characteristics of coordinated strike motion, establish an external ballistic dynamics model for the electromagnetic railgun; An electromagnetic railgun launches an unguided projectile with no maneuverability during flight; control is only possible through initial launch parameters. The launch position is fixed, and multiple projectiles striking an incoming target simultaneously are considered to be moving within the same firing plane. Based on these two points, it is assumed that the unguided projectile launched by the electromagnetic railgun has no lateral maneuvering during flight, and therefore no lateral force or yaw moment. For a given incoming direction angle of... For a single target, the electromagnetic railgun adjusts its muzzle orientation to align all projectiles with the direction of the incoming target. In this configuration, all projectiles fire from the same plane, at the same azimuth angle. The projectile flies within the vertical plane. Therefore, this invention adopts a three-degree-of-freedom rigid body ballistic model within the vertical plane. While fully preserving the dynamic characteristics of the projectile's pitch attitude, it discards the lateral motion and lateral moment-related components, balancing model accuracy and computational efficiency. A ground coordinate system is established with the initial position of the projectile's center of mass at the launch point as the origin. After dimensionality reduction, the projectile only performs translational motion of its center of mass and pitch rotation around the horizontal axis within the vertical plane. Its external ballistic dynamics model for the electromagnetic railgun is as follows: (5) in, The coordinates of the projectile's center of mass in the ground coordinate system within the impact plane; For the projectile's flight time; The velocity of the projectile's center of mass; , These are the trajectory inclination angle and the elevation angle, respectively. It is the pitch angular velocity; For the mass of the projectile; This is the pitch inertia. , , These are the drag, lift components, and pitching moment of aerodynamics in the ballistic coordinate system; specifically: (6) in, For the angle of attack; air density; The reference area of ​​the projectile; The characteristic length of the projectile; This is the drag coefficient; This is the derivative of the lift coefficient with respect to the angle of attack; and These are the derivatives of the pitch moment coefficient with respect to the angle of attack and with respect to the dimensionless pitch velocity, respectively.

[0012] The initial conditions for the external ballistic dynamics model of the electromagnetic railgun are: , , , .

[0013] Step 3-2: Based on the electromagnetic railgun external ballistic dynamics model from Step 3-1, generate a training dataset for training the MDN-MLP concatenated network. The training dataset includes MDN network training samples and MLP network training samples; specifically: Step 3-2-1: Set of candidate initial velocity discrete gears set in step 2 and initial trajectory inclination boundary constraints Uniform random sampling is performed within the defined parameter space to obtain sampling parameters, resulting in several sets of initial parameter combinations. , For the first The initial velocity values ​​corresponding to each gear. For the first The initial trajectory inclination angle of each sample is obtained. For each set of sampled parameters, the fourth-order Runge-Kutta method is used to solve the electromagnetic gun external ballistic dynamics model in step 3-1 to obtain numerical simulation data, obtain the impact point coordinates and projectile flight time corresponding to the impact point, and uniformly sample the trajectory within its flight time range to obtain the sampled dataset.

[0014] Furthermore, to avoid uneven distribution of the sampled dataset along the range direction, the uniform density box principle is used to resample the sampled dataset, ensuring that the sampled dataset is evenly distributed along the range direction. The regions that can be represented are approximately uniformly distributed, thus ensuring that the MDN-MLP cascaded network has good generalization ability in all candidate landing point regions.

[0015] Step 3-2-2: Based on the sampling dataset obtained from numerical simulation data, construct two types of training samples, including MDN network training samples and MLP network training samples; The process of determining the training samples for the MDN network is as follows: the input is the coordinates of the target point. With initial velocity The output is the true value of the initial ballistic inclination angle obtained from sampling. That is, to establish a mapping relationship .in, Indicates the first The coordinates of the landing points of the samples are used to obtain the training samples for the MDN network. The MDN network training samples are used to train the MDN network, enabling it to learn to infer the initial trajectory inclination angle based on the target position and initial velocity.

[0016] The process of determining training samples for an MLP network is as follows: the input is the coordinates of the target point. initial velocity True value of the initial trajectory inclination angle The output is the destination point. Flight time is truly valuable That is, to establish a mapping relationship Finally, the training samples for the MLP network were obtained. The MLP network training samples are used to train the MLP network so that it learns to predict the time required for the projectile to reach the target based on the target's position, initial velocity, and tilt angle.

[0017] Step 3-2-3: Normalize both the MDN network training samples and the MLP network training samples to obtain the normalized MDN network training samples and MLP network training samples, which are the training datasets used for training the MDN-MLP concatenated network.

[0018] Step 4: To overcome the limitations of traditional numerical traversal methods, such as high computational cost and difficulty in fitting the multimodal distribution of the inverse ballistic problem, this invention constructs an MDN-MLP concatenated network based on MDN and MLP networks. Using the training dataset obtained in Step 3, a fast method for solving initial ballistic parameters based on the MDN-MLP concatenated network is established, resulting in a trained MDN-MLP concatenated network. Specifically: Step 4-1: Establish an MDN network to predict the initial ballistic inclination angle and obtain the predicted value of the initial ballistic inclination angle; Hybrid density network, or MDN network for short, uses three neurons as the input layer to receive the first three values ​​from the normalized MDN network training samples in step 3-2. ,use Each layer is a fully connected hidden layer. Each neuron uses ReLU nonlinear activation between layers, outputting a mixing coefficient. mean With log standard deviation The parameters of the Gaussian mixture model. Among them, the number of Gaussian components in the MDN network. The value is set to 2, corresponding to the high-angle and low-angle ballistic trajectories respectively. The MDN network uses the negative log-likelihood loss function (NLL) to fit the multimodal probability distribution. The expression is: (7) in, For training batch size, For batch indexing; Index of Gaussian components; For the first The sample at the th The mixing coefficient over each Gaussian component; and The first The sample at the th Mean and standard deviation of each Gaussian component; For the first The ground truth value of the initial trajectory inclination angle for each sample is the same as the ground truth value of the initial trajectory inclination angle in the training dataset. ; It is the probability density function of the standard normal distribution.

[0019] After training the MDN network using training samples, the mean value of the low-extension trajectory is selected when using the MDN network for prediction. The output is used as the initial trajectory inclination angle prediction value for this trajectory. .

[0020] Step 4-2: Establish an MLP network to predict the flight time of the projectile to the target point and obtain the predicted value of the flight time of the projectile to the target point; A fully connected network, or MLP network for short, uses four neurons as its input layer, receiving the first four values ​​from the normalized MLP network training samples in step 3-2. The hidden layer is layer The nodes are fully connected layers, with ReLU nonlinear activation between layers. The network output layer outputs the predicted flight time of the projectile to the target point. During training, the output layer will receive data from the training samples of the MLP network. MLP networks are trained using the mean squared error loss function (MSE). The expression is: (8) in, For MLP networks to the first The predicted flight time of a projectile to the target point for a sample; For the first The ground truth value of the projectile's time to reach the target point for each sample, compared with the ground truth value of the predicted time to reach the target point for each projectile in the training dataset. ; Step 4-3: Train the MDN network from Step 4-1 and the MLP network from Step 4-2 offline using the Adam optimizer, setting the initial learning rate and batch size. Save the network weights of the trained MDN and MLP networks.

[0021] Step 4-4: After the training processes in Steps 4-1 and 4-2 are completed offline, when using it online, only the network weights from Step 4-3 need to be loaded to construct the MDN-MLP concatenated network.

[0022] When using it online, first enter through the MDN network. Output at Initial velocity reached Predicted initial trajectory inclination angle for low-profile ballistics Then input through the MLP network Solve for the initial trajectory inclination angle The corresponding predicted flight time to the target point The predicted initial trajectory inclination angle and the predicted flight time of the projectile to the target point constitute the trajectory parameters, which serve as the prediction results.

[0023] Step 5: The MDN-MLP concatenated network trained in Step 4 can quickly provide predicted values ​​for the initial trajectory inclination angle and flight time to the target point, but its predicted impact point often has residual bias. To improve the accuracy of the solution, this invention designs a geometric correction algorithm based on direction error feedback. The geometric correction algorithm utilizes the angular relationship between the deviation vector between the predicted impact point and the target point, and the velocity vector at the projectile's terminal velocity, combined with the current inclination angle's position in the ascent or descent phase of the trajectory, to adaptively determine the correction direction and magnitude for the inclination angle and flight time. The geometric correction algorithm receives the prediction results output from Step 4, iteratively corrects them, and outputs parameters that satisfy the impact point accuracy constraints. Specifically: Step 5-1: Initialize the parameters of the geometric correction algorithm based on direction error feedback, and calculate the deviation vector and direction difference angle.

[0024] make Set the maximum number of iterations. .

[0025] Assuming that when using an MDN-MLP concatenated network, the input is the real target point. With a given initial velocity The ballistic parameters to be corrected are obtained as the predicted values ​​of the initial ballistic inclination angle. Compared with flight time prediction Using the electromagnetic railgun external ballistic dynamics model from step 3-1, integrate to... At that moment, the predicted landing point was obtained. And the projectile's terminal velocity vector at that moment is .in These represent the horizontal distance and height of the actual target point, respectively. These represent the horizontal distance and height of the predicted landing point, respectively. These are the horizontal and vertical components of the terminal velocity, respectively.

[0026] Define from the predicted landing point point to deviation vector for: (9) Current landing point error for: (10) in, , They are respectively The deviation component in the direction.

[0027] Calculate the deviation vector With the terminal velocity vector Direction difference angle for: (11) in, The four-quadrant arctangent function returns a value in degrees.

[0028] Will Standardize the quadrant transformation operation. Within the interval. If Corresponding landing point error Less than the preset threshold error If the correction is skipped, the initial parameters of the single-shot trajectory predicted by the MDN-MLP concatenated network will be directly applied. The first solution is directly output as a candidate solution for a single-shot trajectory; otherwise, it needs to be corrected, proceeding to step 5-2. Therefore, the initial parameters for the single-shot trajectory in the 0th iteration are: The preset threshold error The range is .

[0029] Step 5-2: When determining the need for correction in Step 5-1, determine whether the direction of angle correction needs to be reversed based on the vertical component of the initial trajectory inclination angle and the terminal velocity. That is, this invention introduces a large-angle reversal rule to address the different characteristics of high-angle and low-angle projectile trajectories. Specifically: Based on the initial ballistic inclination angle prediction value obtained in step 5-1 The vertical component of the current terminal velocity is Determine the angle sign factor. When and At this point, it indicates that the projectile is already in the descent phase of a high-angle trajectory. Increasing the angle at this stage actually brings the impact point closer, while decreasing the angle actually makes the impact point farther away, which is the opposite of the characteristics of a low-angle trajectory. Therefore, an angle sign factor is defined. In order to satisfy and Time to take In other cases, take To achieve a global reversal of the angle correction direction, where This is a dimensionless sign factor. The large tilt angle reversal rule ensures that the algorithm has the correct correction direction throughout the entire high and low trajectory.

[0030] Step 5-3: Confirm after performing step 5-2. Then, the angle correction sign and time correction sign are determined according to the quadrant in which the direction difference angle is located, and the correction magnitude, including the angle correction amount and the time correction amount, is calculated.

[0031] Step 5-1 has been completed Standardized to The interval, therefore according to Determine the angle correction sign within the specified interval. and time correction symbols Both are dimensionless symbolic factors. At that time, take ;when At that time, take ;when At that time, take ;when At that time, take ; Define the landing point sensitivity coefficient ,in The coefficient is dimensionless. is the base of the natural logarithm, approximately 2.718; For the first The initial trajectory inclination angle is obtained from the nth iteration. The impact point sensitivity coefficient decreases as the initial trajectory inclination angle increases, reflecting the sensitivity of the impact point to the vertical deviation of a unit change in the initial trajectory inclination angle. Then the nth iteration... Angle correction amount in the next iteration and time correction amount They are respectively: (12) (13) in, For the first Error at the landing point of the next iteration; For the first Next iteration direction difference angle; For the first The next iteration predicts the height of the landing point; For the first The horizontal component of the velocity at the end of the next iteration; For the first The vertical component of the velocity at the end of the next iteration; add 1.0 to the denominator to prevent division by zero anomalies.

[0032] Step 5-4: After determining the correction magnitude for each iteration in Step 5-3, update the initial parameters of the single-shot trajectory and treat them as candidate parameters. Finally, check whether the candidate parameters meet the impact accuracy constraint, and save the candidate parameters that meet the impact accuracy constraint. The candidate parameters are calculated as follows: (14) in, These are the updated initial trajectory inclination and flight time to the target point, respectively. For the first The flight time of the projectile to the target point is obtained from the next iteration.

[0033] Will Forced constraints to physical candidate domains , For candidate parameters Recalculate the ballistic integral and recalculate the new error. .like Then accept candidate parameters. ;like Reject candidate parameters Continue iterating; Step 5-5: Continue the iterations from Steps 5-2 to 5-4. If the landing point error satisfies... The correction is then deemed successful, and the candidate parameters are... Considered as a candidate solution for single-shot ballistics If the accuracy requirement is not met even after reaching the maximum number of iterations, then the candidate parameters are discarded. These are marked as unavailable candidate solutions. Finally, a geometric correction algorithm is used to obtain corrected single-shot trajectory candidate solutions that satisfy the impact point accuracy constraint, which are then passed to step 6 to generate a candidate solution set for hitting the target point.

[0034] Step 6: Step 5 geometrically corrected the predicted initial trajectory inclination angle and the predicted flight time of the projectile to the target point for each initial velocity discrete level, obtaining single-shot trajectory candidate solutions that satisfy the impact accuracy constraint. However, the electromagnetic railgun is limited by the initial velocity discrete level constraint in Step 2 and the minimum firing cycle of the gun, so not all single-shot trajectory candidate solutions can form an executable coordinated attack sequence at the same expected hit time. Therefore, this invention further establishes a multi-projectile coordinated attack sequence planning rule to select coordinated attack sequences that satisfy the firing interval constraint from the corrected single-shot trajectory candidate solutions obtained in Step 5.

[0035] The multi-missile coordinated strike sequence planning rule comprises two steps. First, it iterates through all discrete initial velocity levels, applying the MDN-MLP concatenated network from step 4 and the geometric correction algorithm from step 5 for each level to generate all single-shot trajectory candidate solutions that satisfy the impact point accuracy constraint, forming a candidate solution set for hitting the target point. Then, a greedy rule aimed at maximizing the number of coordinated sequences is used to select sequences from the candidate solution set that satisfy the adjacent launch time interval constraint, considering these as candidate coordinated strike sequences. Finally, the initial velocity, initial trajectory inclination angle, launch delay time, and expected hit time of each projectile are output. Specifically: Step 6-1: Generate a candidate solution set for hitting the target point.

[0036] For a given target point and expected hit time Traverse the discrete gear set of initial velocity For each discrete initial velocity level First, call the trained MDN-MLP concatenated network from step 4 and input the target point coordinates. and initial velocity Quickly obtain the initial ballistic inclination angle prediction value and the projectile's flight time to the target point prediction value. .

[0037] Then, the geometric correction algorithm based on direction error feedback from step 5 is used for iterative correction. If the landing point error is corrected... This yields a candidate solution for a single-shot ballistic trajectory. ,in For the first A corrected initial trajectory inclination angle For the first The corrected flight time of the projectile to reach the target point. Calculate the... Candidate solution launch waiting time That is, the time required from the current moment until the launch of the projectile, where, The expected hit time. If Then the candidate solutions for single-shot ballistics will be retained.

[0038] After traversing all discrete initial velocity levels, all retained single-shot trajectory candidate solutions constitute the target hit point candidate solution set. ,in This represents the number of candidate solutions for a single-shot ballistic trajectory. Indicates the first A single-shot ballistic candidate solution This is the first candidate solution for a single-shot trajectory. This is the second candidate solution for a single-shot trajectory. For the first One single-shot ballistic candidate solution.

[0039] Step 6-2: Construct a cooperative strike sequence generation algorithm based on the candidate solution set of hit target points obtained in Step 6-1.

[0040] The candidate solution set of hit target points obtained in step 6-1 All candidate solutions are ranked according to flight time. The order is ascending, so the shorter the flight time, the higher the ranking. Let the minimum firing period of the electromagnetic railgun be... , The minimum time interval required between the launch times of two adjacent projectiles is determined by the physical characteristics of the artillery's loading and resetting mechanism. This invention employs a greedy rule aimed at maximizing the number of cooperative sequences to select candidate cooperative strike sequences that satisfy the launch interval constraint from the candidate solution set of target hit points, i.e., designing a cooperative strike sequence generation algorithm.

[0041] The cooperative attack sequence generation algorithm first initializes the cooperative attack sequence. Given an empty sequence, iterate through the sorted candidate solution set for hitting the target point. In ,like If empty, then Join the coordinated strike sequence .like If the value is not empty, it proves that there exists a single-shot ballistic candidate solution in the sequence. The last single-shot ballistic candidate solution added in the middle is , The corresponding launch waiting time is Then the first Launch latency of a single-shot ballistic candidate solution .in, For the last time to join the coordinated strike sequence The initial velocity of the candidate solution for the single-shot trajectory. For the last time to join the coordinated strike sequence The initial trajectory inclination angle after correction of the candidate solution for a single-shot ballistic trajectory. For the last time to join the coordinated strike sequence The flight time of a single-shot ballistic candidate solution after correction to the projectile's arrival at the target point.

[0042] If satisfied Then the current single-shot ballistic candidate solution will be... Join the coordinated strike sequence and update Otherwise, skip the current single-shot trajectory candidate solution and continue iterating through subsequent single-shot trajectory candidate solutions. For the last time joined The launch latency of a single-shot ballistic candidate solution; This is the minimum firing cycle for the electromagnetic railgun.

[0043] After the traversal is complete, the coordinated strike sequence This refers to a candidate coordinated strike sequence that satisfies both the simultaneous hit and launch interval constraints. If If empty, it indicates that a coordinated strike plan cannot be planned under the current initial velocity dispersion level and range limitations, and adjustments are needed. Alternatively, consider extending the discrete initial velocity settings. Ultimately, the candidate coordinated strike sequence will be considered as a coordinated strike sequence. In a coordinated strike sequence, each projectile contains a defined initial launch velocity. Initial ballistic inclination angle Launch waiting time and expected hit time , as the direct command for the fire control of the electromagnetic gun.

[0044] The overall execution flow of the method of this invention is as follows: Upon receiving the coordinated strike mission, the MDN-MLP concatenated network model trained offline in step 4 is first loaded. Then, step 6-1 is executed, sequentially calling the MDN-MLP concatenated network model and the geometric correction algorithm from step 5 for each discrete initial velocity level to generate a candidate solution set. Next, the coordinated strike sequence generation algorithm from step 6-2 is executed to select the coordinated strike sequence that satisfies the launch interval constraint from the candidate solutions. Finally, the coordinated strike sequence is output. Each projectile contains a defined initial velocity, initial trajectory angle, firing delay time, and expected hit time, serving as direct commands for the electromagnetic gun's fire control.

[0045] The beneficial effects of this invention are as follows: (1) This invention addresses the multimodal distribution characteristics of high-angle and low-angle trajectories corresponding to the same target point in the inverse ballistic problem, as well as the problem that the computational cost of traditional numerical traversal methods increases exponentially with increasing resolution, making it difficult to meet the real-time requirements of rapid strikes. It designs a fast solution method for initial ballistic parameters by cascading a hybrid density network (MDN) and a fully connected network (MLP). The MDN network takes the target point coordinates and initial velocity as input and uses two Gaussian components to fit the multimodal probability distribution of the inverse ballistic problem, directly outputting the predicted initial trajectory inclination angle of the low-angle trajectory. The MLP network takes this predicted value as input and outputs the corresponding flight time. After offline training, this method only requires two forward propagations to obtain the initial trajectory parameters during online solving, eliminating the need for iterative search. The single solution time reaches the millisecond level, improving computational efficiency by several orders of magnitude compared to traditional numerical traversal methods. This fundamentally solves the technical problems of low solution efficiency and difficulty in fitting multimodal distributions in the inverse ballistic problem.

[0046] (2) To address the problem that neural networks often exhibit large prediction errors in sparse training data regions or at ballistic boundaries, making it difficult to directly meet the high-precision requirements of coordinated strikes, this invention designs a geometric correction algorithm based on directional error feedback. This algorithm does not rely on large-scale additional data or complex optimization solutions. Instead, it uses the directional difference angle between the deviation vector and the terminal velocity vector to determine the dominant source of the impact point error and determines the correction direction based on the quadrant where the directional difference angle is located. Simultaneously, a large-angle reversal rule is introduced to solve the special problem of opposite angle correction directions during the descent phase of a high-altitude ballistic trajectory. Through an iterative strategy with adaptive step size adjustment and random perturbation, this algorithm typically requires only 25 iterations to converge the impact point error from over 1,000 meters to within 60 meters, achieving a balance between prediction accuracy and solution speed, enabling the neural network prediction results to meet engineering impact point accuracy requirements.

[0047] (3) This invention addresses the problems of uncontrolled projectiles being unable to coordinate online, requiring parameter planning before launch, and the physical constraints of artillery firing intervals. It designs a greedy coordination rule with the goal of maximizing the number of coordination sequences. This rule first iterates through discrete initial velocity levels to generate a candidate solution set of single-shot trajectories that satisfy the constraints of impact accuracy and flight time. Then, it sorts these solutions in ascending order of flight time and performs greedy filtering with the minimum firing period as a constraint. Within polynomial time, it generates an executable coordinated strike sequence that satisfies the requirement that all projectiles simultaneously hit the same target and that the time interval between adjacent launches satisfies the physical limitations of the artillery. This solves the problem that traditional coordination methods, which assume full controllability and inter-projectile information interaction, are unsuitable for completely uncontrolled electromagnetic railgun projectiles. Each projectile output contains a clear initial launch velocity, initial trajectory inclination, launch time, and expected hit time, which can be directly executed as a command for electromagnetic railgun fire control.

[0048] In summary, through the synergistic effect of the three innovative aspects mentioned above, this invention has for the first time achieved coordinated planning for multiple simultaneous hits of uncontrolled projectiles from an electromagnetic railgun. While ensuring engineering accuracy, it meets the real-time requirements of rapid strike missions and provides a candidate technical solution for electromagnetic railguns to intercept high-speed maneuvering targets. Attached Figure Description

[0049] Figure 1 This is a flowchart of the present invention.

[0050] Figure 2 This is a schematic diagram of the trajectory of the electromagnetic railgun coordinating with the target point (5000, 10000) of the present invention.

[0051] Figure 3 The example is a cooperative strike trajectory diagram for example (5000, 10000).

[0052] Figure 4 The collaborative strike Gantt chart planned for the example (5000, 10000).

[0053] Figure 5 The example is a cooperative strike trajectory diagram for example (20000, 3000).

[0054] Figure 6 The collaborative strike Gantt chart planned for the example (20000, 3000). Detailed Implementation

[0055] The present invention will be further described below with reference to specific embodiments.

[0056] For two types of target points (5000, 10000) and (5000, 10000) as examples, a coordinated strike mission is performed. The electromagnetic railgun is deployed at the origin of the ground inertial coordinate system, where the two types of target points are located at... Figure 2Strike plane The coordinates of the point. Figure 1 This is an overall flowchart of the method of the present invention. Figure 2 This is a schematic diagram illustrating the coordinated attack of electromagnetic railguns on target points. Using a ground coordinate system, Represents the initial trajectory inclination angles of two different ballistic trajectories. The target point is [location]. All multiple missiles fly within a single impact plane. The following steps are used to plan a coordinated multi-missile strike to achieve a simultaneous hit: Step 1: The solution process for the electromagnetic railgun multi-projectile coordinated attack planning problem is divided into two sub-problems: accurately solving for the initial parameters of a single-shot trajectory and generating a multi-projectile coordinated attack sequence; specifically: Electromagnetic railguns utilize the advantages of high initial velocity, long range, fast response, low cost, and strong controllability of firing parameters to rapidly launch multiple projectiles that arrive at the target area simultaneously, forming an instantaneous barrage, thereby significantly increasing the probability of destroying high-speed targets.

[0057] The solution to the electromagnetic railgun multi-projectile coordinated strike planning problem is as follows: First, given the target point, the desired hit time, and the physical constraints of the gun, the inverse external ballistic dynamics problem is solved quickly to obtain the initial launch velocity, initial trajectory inclination angle, and flight time of each projectile to the target point. Then, the trajectory parameters that meet the impact accuracy constraints are selected from multiple trajectories, and an executable coordinated strike sequence is generated to ensure that all projectiles hit the target synchronously within the allowed time window.

[0058] This invention divides the solution process for the multi-projectile coordinated strike planning problem of electromagnetic railguns into two sub-problems: the first is the rapid and accurate solution of the initial parameters of a single-shot trajectory, referred to as the single-shot solution, where the initial trajectory parameters of a single-shot trajectory include the initial trajectory inclination angle of the projectile and the flight time of the projectile to reach the target point; the second is the generation of the multi-projectile coordinated strike sequence. The single-shot solution further includes the prediction of multimodal parameters of the inverse trajectory problem and the geometric correction of residual errors.

[0059] Step 2: Set the physical constraints that must be met when solving the initial ballistic parameters to ensure the feasibility of the solution in actual engineering.

[0060] When solving for initial trajectory parameters to plan coordinated attacks or projectile launches, electromagnetic railguns must satisfy the physical constraints proposed in this invention, including initial velocity discrete step constraints, initial trajectory inclination angle boundary constraints, trajectory rationality constraints, and impact point accuracy constraints. Specifically: The initial velocity discrete range constraint: The initial velocity of the electromagnetic railgun is determined by hardware conditions such as power supply energy and track length, and cannot be continuously and steplessly adjusted. In practical engineering, only a limited number of discrete ranges can be selected. Therefore, during parameter planning, the initial velocity value must be constrained to belong to the candidate discrete initial velocity range set. Let the candidate discrete initial velocity range set be... for: (1) in, This indicates the number of gears, which is determined by the actual performance of the electromagnetic gun. For the first The initial velocity value corresponding to each gear; therefore This is the initial velocity value corresponding to the first gear; This is the initial velocity value corresponding to the second gear; For the first The initial velocity value corresponding to each gear.

[0061] The initial trajectory inclination angle boundary constraint: the initial trajectory inclination angle of the projectile's outer trajectory. The elevation angle limit of the electromagnetic railgun launching mechanism needs to be met, and its range is set as follows: (2) in, The initial trajectory angle of the projectile; and These are the minimum and maximum pitch angles allowed by the electromagnetic railgun launching mechanism, respectively.

[0062] The ballistic rationality constraint is as follows: Due to the dimensionality reduction simplification achieved by employing the small angle-of-attack assumption in the external ballistic dynamics model, excessive angle-of-attack or its rate of change during flight will lead to significant lift nonlinearity and drastic changes in the moment coefficient, causing the model to fail and the solution results to lose physical meaning. Therefore, an upper limit constraint needs to be imposed on the projectile's angle of attack and its rate of change. (3) in, For the angle of attack of the projectile; Let be the rate of change of the projectile's angle of attack; and These are the set upper limit for angle of attack and the upper limit for rate of change of angle of attack, respectively.

[0063] The impact point accuracy constraint is as follows: to ensure the effectiveness of the strike, the Euclidean distance between the impact point and the target point corresponding to the calculated initial ballistic parameters must be less than a preset threshold. ,Right now: (4) in, The point of impact corresponding to the calculated ballistic parameters; For the target point; The preset threshold value varies depending on the type of propellant used in the projectile. .

[0064] Step 3: After clarifying the physical constraints in Step 2, in order to obtain accurate projectile flight trajectory and time information, an external ballistic dynamics model is established based on the flight motion characteristics of the uncontrolled projectile, and a training dataset for training the MDN-MLP concatenated network is constructed accordingly; specifically: Step 3-1: Based on the characteristics of coordinated strike motion, establish an external ballistic dynamics model for the electromagnetic railgun; An electromagnetic railgun launches an unguided projectile with no maneuverability during flight; control is only possible through initial launch parameters. The launch position is fixed, and multiple projectiles striking an incoming target simultaneously are considered to be moving within the same firing plane. Based on these two points, it is assumed that the unguided projectile launched by the electromagnetic railgun has no lateral maneuvering during flight, and therefore no lateral force or yaw moment. For a given incoming direction angle of... For a single target, the electromagnetic railgun adjusts its muzzle orientation to align all projectiles with the direction of the incoming target. In this configuration, all projectiles fire from the same plane, at the same azimuth angle. The projectile flies within the vertical plane. Therefore, this invention adopts a three-degree-of-freedom rigid body ballistic model within the vertical plane. While fully preserving the dynamic characteristics of the projectile's pitch attitude, it discards the lateral motion and lateral moment-related components, balancing model accuracy and computational efficiency. A ground coordinate system is established with the initial position of the projectile's center of mass at the launch point as the origin. After dimensionality reduction, the projectile only performs translational motion of its center of mass and pitch rotation around the horizontal axis within the vertical plane. Its external ballistic dynamics model for the electromagnetic railgun is as follows: (5) in, The coordinates of the projectile's center of mass in the ground coordinate system within the impact plane; For the projectile's flight time; The velocity of the projectile's center of mass; , These are the trajectory inclination angle and the elevation angle, respectively. It is the pitch angular velocity; For the mass of the projectile; This is the pitch inertia. , , These are the drag, lift components, and pitching moment of aerodynamics in the ballistic coordinate system; specifically: (6) in, For the angle of attack; air density; The reference area of ​​the projectile; The characteristic length of the projectile; This is the drag coefficient; This is the derivative of the lift coefficient with respect to the angle of attack; and These are the derivatives of the pitch moment coefficient with respect to the angle of attack and with respect to the dimensionless pitch velocity, respectively.

[0065] The initial conditions for the external ballistic dynamics model of the electromagnetic railgun are: , , , .

[0066] Step 3-2: Based on the electromagnetic railgun external ballistic dynamics model from Step 3-1, generate a training dataset for training the MDN-MLP concatenated network. The training dataset includes MDN network training samples and MLP network training samples; specifically: Step 3-2-1: Set of candidate initial velocity discrete gears set in step 2 and initial trajectory inclination boundary constraints Uniform random sampling is performed within the defined parameter space to obtain sampling parameters, resulting in several sets of initial parameter combinations. , For the first The initial velocity values ​​corresponding to each gear. For the first The initial trajectory inclination angle of each sample is obtained. For each set of sampled parameters, the fourth-order Runge-Kutta method is used to solve the electromagnetic gun external ballistic dynamics model in step 3-1 to obtain numerical simulation data, obtain the impact point coordinates and projectile flight time corresponding to the impact point, and uniformly sample the trajectory within its flight time range to obtain the sampled dataset.

[0067] Furthermore, to avoid uneven distribution of the sampled dataset along the range direction, the uniform density box principle is used to resample the sampled dataset, ensuring that the sampled dataset is evenly distributed along the range direction. The regions that can be represented are approximately uniformly distributed, thus ensuring that the MDN-MLP cascaded network has good generalization ability in all candidate landing point regions.

[0068] Step 3-2-2: Based on the sampling dataset obtained from numerical simulation data, construct two types of training samples, including MDN network training samples and MLP network training samples; The process of determining the training samples for the MDN network is as follows: the input is the coordinates of the target point. With initial velocity The output is the true value of the initial ballistic inclination angle obtained from sampling. That is, to establish a mapping relationship .in, Indicates the first The coordinates of the landing points of the samples are used to obtain the training samples for the MDN network. The MDN network training samples are used to train the MDN network, enabling it to learn to infer the initial trajectory inclination angle based on the target position and initial velocity.

[0069] The process of determining training samples for an MLP network is as follows: the input is the coordinates of the target point. initial velocity True value of the initial trajectory inclination angle The output is the destination point. Flight time is truly valuable That is, to establish a mapping relationship Finally, the training samples for the MLP network were obtained. The MLP network training samples are used to train the MLP network so that it learns to predict the time required for the projectile to reach the target based on the target's position, initial velocity, and tilt angle.

[0070] Step 3-2-3: Normalize both the MDN network training samples and the MLP network training samples to obtain the normalized MDN network training samples and MLP network training samples, which are the training datasets used for training the MDN-MLP concatenated network.

[0071] Step 4: To overcome the limitations of traditional numerical traversal methods, such as high computational cost and difficulty in fitting the multimodal distribution of the inverse ballistic problem, this invention constructs an MDN-MLP concatenated network based on MDN and MLP networks. Using the training dataset obtained in Step 3, a fast method for solving initial ballistic parameters based on the MDN-MLP concatenated network is established, resulting in a trained MDN-MLP concatenated network. Specifically: Step 4-1: Establish an MDN network to predict the initial ballistic inclination angle and obtain the predicted value of the initial ballistic inclination angle; Hybrid density network, or MDN network for short, uses three neurons as the input layer to receive the first three values ​​from the normalized MDN network training samples in step 3-2. ,use Each layer is a fully connected hidden layer. Each neuron uses ReLU nonlinear activation between layers, outputting a mixing coefficient. mean With log standard deviation The parameters of the Gaussian mixture model. Among them, the number of Gaussian components in the MDN network. The value is set to 2, corresponding to the high-angle and low-angle ballistic trajectories respectively. The MDN network uses the negative log-likelihood loss function (NLL) to fit the multimodal probability distribution. The expression is: (7) in, For training batch size, For batch indexing; Index of Gaussian components; For the first The sample at the th The mixing coefficient over each Gaussian component; and The first The sample at the th Mean and standard deviation of each Gaussian component; For the first The ground truth value of the initial trajectory inclination angle for each sample is the same as the ground truth value of the initial trajectory inclination angle in the training dataset. ; It is the probability density function of the standard normal distribution.

[0072] After training the MDN network using training samples, the mean value of the low-extension trajectory is selected when using the MDN network for prediction. The output is used as the initial trajectory inclination angle prediction value for this trajectory. .

[0073] Step 4-2: Establish an MLP network to predict the flight time of the projectile to the target point and obtain the predicted value of the flight time of the projectile to the target point; A fully connected network, or MLP network for short, uses four neurons as its input layer, receiving the first four values ​​from the normalized MLP network training samples in step 3-2. The hidden layer is layer The nodes are fully connected layers, with ReLU nonlinear activation between layers. The network output layer outputs the predicted flight time of the projectile to the target point. During training, the output layer will receive data from the training samples of the MLP network. MLP networks are trained using the mean squared error loss function (MSE). The expression is: (8) in, For MLP networks to the first The predicted flight time of a projectile to the target point for a sample; For the first The ground truth value of the projectile's time to reach the target point for each sample, compared with the ground truth value of the predicted time to reach the target point for each projectile in the training dataset. ; Step 4-3: Train the MDN network from Step 4-1 and the MLP network from Step 4-2 offline using the Adam optimizer. Set the initial learning rate to 0.001 and the batch size to 256, and train for a total of 5000 epochs. Save the network weights of the trained MDN and MLP networks.

[0074] Step 4-4: After the training processes in Steps 4-1 and 4-2 are completed offline, when using it online, only the network weights from Step 4-3 need to be loaded to construct the MDN-MLP concatenated network.

[0075] When using it online, first enter through the MDN network. Output at Initial velocity reached Predicted initial trajectory inclination angle for low-profile ballistics Then input through the MLP network Solve for the initial trajectory inclination angle The corresponding predicted flight time to the target point The predicted initial trajectory inclination angle and the predicted flight time of the projectile to the target point constitute the trajectory parameters, which serve as the prediction results.

[0076] Step 5: The MDN-MLP concatenated network trained in Step 4 can quickly provide predicted values ​​for the initial trajectory inclination angle and flight time to the target point, but its predicted impact point often has residual bias. To improve the accuracy of the solution, this invention designs a geometric correction algorithm based on direction error feedback. The geometric correction algorithm utilizes the angular relationship between the deviation vector between the predicted impact point and the target point, and the velocity vector at the projectile's terminal velocity, combined with the current inclination angle's position in the ascent or descent phase of the trajectory, to adaptively determine the correction direction and magnitude for the inclination angle and flight time. The geometric correction algorithm receives the prediction results output from Step 4, iteratively corrects them, and outputs parameters that satisfy the impact point accuracy constraints. Specifically: Step 5-1: Initialize the parameters of the geometric correction algorithm based on direction error feedback, and calculate the deviation vector and direction difference angle.

[0077] make Set the maximum number of iterations. .

[0078] Assuming that when using an MDN-MLP concatenated network, the input is the real target point. With a given initial velocity The ballistic parameters to be corrected are obtained as the predicted values ​​of the initial ballistic inclination angle. Compared with flight time prediction Using the electromagnetic railgun external ballistic dynamics model from step 3-1, integrate to... At that moment, the predicted landing point was obtained. And the projectile's terminal velocity vector at that moment is .in These represent the horizontal distance and height of the actual target point, respectively. These represent the horizontal distance and height of the predicted landing point, respectively. These are the horizontal and vertical components of the terminal velocity, respectively.

[0079] Define from the predicted landing point point to deviation vector for: (9) Current landing point error for: (10) in, , They are respectively The deviation component in the direction.

[0080] Calculate the deviation vector With the terminal velocity vector Direction difference angle for: (11) in, The four-quadrant arctangent function returns a value in degrees.

[0081] Will Standardize the quadrant transformation operation. Within the interval. If Corresponding landing point error Less than the preset threshold error If the correction is skipped, the initial parameters of the single-shot trajectory predicted by the MDN-MLP concatenated network will be directly applied. The first solution is directly output as a candidate solution for a single-shot trajectory; otherwise, it needs to be corrected, proceeding to step 5-2. Therefore, the initial parameters for the single-shot trajectory in the 0th iteration are: The preset threshold error The range is .

[0082] Step 5-2: When determining the need for correction in Step 5-1, determine whether the direction of angle correction needs to be reversed based on the vertical component of the initial trajectory inclination angle and the terminal velocity. That is, this invention introduces a large-angle reversal rule to address the different characteristics of high-angle and low-angle projectile trajectories. Specifically: Based on the initial ballistic inclination angle prediction value obtained in step 5-1 The vertical component of the current terminal velocity is Determine the angle sign factor. When and At this point, it indicates that the projectile is already in the descent phase of a high-angle trajectory. Increasing the angle at this stage actually brings the impact point closer, while decreasing the angle actually makes the impact point farther away, which is the opposite of the characteristics of a low-angle trajectory. Therefore, an angle sign factor is defined. In order to satisfy and Time to take In other cases, take To achieve a global reversal of the angle correction direction, where This is a dimensionless sign factor. The large tilt angle reversal rule ensures that the algorithm has the correct correction direction throughout the entire high and low trajectory.

[0083] Step 5-3: Confirm after performing step 5-2. Then, the angle correction sign and time correction sign are determined according to the quadrant in which the direction difference angle is located, and the correction magnitude, including the angle correction amount and the time correction amount, is calculated.

[0084] Step 5-1 has been completed Standardized to The interval, therefore according to Determine the angle correction sign within the specified interval. and time correction symbols Both are dimensionless symbolic factors. At that time, take ;when At that time, take ;when At that time, take ;when At that time, take ; Define the landing point sensitivity coefficient ,in The coefficient is dimensionless. is the base of the natural logarithm, approximately 2.718; For the first The initial trajectory inclination angle is obtained from the nth iteration. The impact point sensitivity coefficient decreases as the initial trajectory inclination angle increases, reflecting the sensitivity of the impact point to the vertical deviation of a unit change in the initial trajectory inclination angle. Then the nth iteration... Angle correction amount in the next iteration and time correction amount They are respectively: (12) (13) in, For the first Error at the landing point of the next iteration; For the first Next iteration direction difference angle; For the first The next iteration predicts the height of the landing point; For the first The horizontal component of the velocity at the end of the next iteration; For the first The vertical component of the velocity at the end of the next iteration; add 1.0 to the denominator to prevent division by zero anomalies.

[0085] Step 5-4: After determining the correction magnitude for each iteration in Step 5-3, update the initial parameters of the single-shot trajectory and treat them as candidate parameters. Finally, check whether the candidate parameters meet the impact accuracy constraint, and save the candidate parameters that meet the impact accuracy constraint. The candidate parameters are calculated as follows: (14) in, These are the updated initial trajectory inclination and flight time to the target point, respectively. For the first The flight time of the projectile to the target point is obtained from the next iteration.

[0086] Will Forced constraints to physical candidate domains , For candidate parameters Recalculate the ballistic integral and recalculate the new error. .like Then accept candidate parameters. ;like Reject candidate parameters Continue iterating; Step 5-5: Continue the iterations from Steps 5-2 to 5-4. If the landing point error satisfies... The correction is then deemed successful, and the candidate parameters are... Considered as a candidate solution for single-shot ballistics If the accuracy requirement is not met even after reaching the maximum number of iterations, then the candidate parameters are discarded. These are marked as unavailable candidate solutions. Finally, a geometric correction algorithm is used to obtain corrected single-shot trajectory candidate solutions that satisfy the impact point accuracy constraint, which are then passed to step 6 to generate a candidate solution set for hitting the target point.

[0087] Step 6: Step 5 geometrically corrected the predicted initial trajectory inclination angle and the predicted flight time of the projectile to the target point for each initial velocity discrete level, obtaining single-shot trajectory candidate solutions that satisfy the impact accuracy constraint. However, the electromagnetic railgun is limited by the initial velocity discrete level constraint in Step 2 and the minimum firing cycle of the gun, so not all single-shot trajectory candidate solutions can form an executable coordinated attack sequence at the same expected hit time. Therefore, this invention further establishes a multi-projectile coordinated attack sequence planning rule to select coordinated attack sequences that satisfy the firing interval constraint from the corrected single-shot trajectory candidate solutions obtained in Step 5.

[0088] The multi-missile coordinated strike sequence planning rule comprises two steps. First, it iterates through all discrete initial velocity levels, applying the MDN-MLP concatenated network from step 4 and the geometric correction algorithm from step 5 for each level to generate all single-shot trajectory candidate solutions that satisfy the impact point accuracy constraint, forming a candidate solution set for hitting the target point. Then, a greedy rule aimed at maximizing the number of coordinated sequences is used to select sequences from the candidate solution set that satisfy the adjacent launch time interval constraint, considering these as candidate coordinated strike sequences. Finally, the initial velocity, initial trajectory inclination angle, launch delay time, and expected hit time of each projectile are output. Specifically: Step 6-1: Generate a candidate solution set for hitting the target point.

[0089] For a given target point and expected hit time Traverse the discrete gear set of initial velocity For each discrete initial velocity level First, call the trained MDN-MLP concatenated network from step 4 and input the target point coordinates. and initial velocity Quickly obtain the initial ballistic inclination angle prediction value and the projectile's flight time to the target point prediction value. .

[0090] Then, the geometric correction algorithm based on direction error feedback from step 5 is used for iterative correction. If the landing point error is corrected... This yields a candidate solution for a single-shot ballistic trajectory. ,in For the first A corrected initial trajectory inclination angle For the first The corrected flight time of the projectile to reach the target point. Calculate the... Candidate solution launch waiting time That is, the time required from the current moment until the launch of the projectile, where, The expected hit time. If Then the candidate solutions for single-shot ballistics will be retained.

[0091] After traversing all discrete initial velocity levels, all retained single-shot trajectory candidate solutions constitute the target hit point candidate solution set. ,in This represents the number of candidate solutions for a single-shot ballistic trajectory. Indicates the first A single-shot ballistic candidate solution This is the first candidate solution for a single-shot trajectory. This is the second candidate solution for a single-shot trajectory. For the first One single-shot ballistic candidate solution.

[0092] Step 6-2: Construct a cooperative strike sequence generation algorithm based on the candidate solution set of hit target points obtained in Step 6-1.

[0093] The candidate solution set of hit target points obtained in step 6-1 All candidate solutions are ranked according to flight time. The order is ascending, so the shorter the flight time, the higher the ranking. Let the minimum firing period of the electromagnetic railgun be... , The minimum time interval required between the launch times of two adjacent projectiles is determined by the physical characteristics of the artillery's loading and resetting mechanism. This invention employs a greedy rule aimed at maximizing the number of cooperative sequences to select candidate cooperative strike sequences that satisfy the launch interval constraint from the candidate solution set of target hit points, i.e., designing a cooperative strike sequence generation algorithm.

[0094] The cooperative attack sequence generation algorithm first initializes the cooperative attack sequence. Given an empty sequence, iterate through the sorted candidate solution set for hitting the target point. In ,like If empty, then Join the coordinated strike sequence .like If the value is not empty, it proves that there exists a single-shot ballistic candidate solution in the sequence. The last single-shot ballistic candidate solution added in the middle is , The corresponding launch waiting time is Then the first Launch latency of a single-shot ballistic candidate solution .in, For the last time to join the coordinated strike sequence The initial velocity of the candidate solution for the single-shot trajectory. For the last time to join the coordinated strike sequence The initial trajectory inclination angle after correction of the candidate solution for a single-shot ballistic trajectory. For the last time to join the coordinated strike sequence The flight time of a single-shot ballistic candidate solution after correction to the projectile's arrival at the target point.

[0095] If satisfied Then the current single-shot ballistic candidate solution will be... Join the coordinated strike sequence and update Otherwise, skip the current single-shot trajectory candidate solution and continue iterating through subsequent single-shot trajectory candidate solutions. For the last time joined The launch latency of a single-shot ballistic candidate solution; This is the minimum firing cycle for the electromagnetic railgun.

[0096] After the traversal is complete, the coordinated strike sequence This refers to a candidate coordinated strike sequence that satisfies both the simultaneous hit and launch interval constraints. If If empty, it indicates that a coordinated strike plan cannot be planned under the current initial velocity dispersion level and range limitations, and adjustments are needed. Alternatively, consider extending the discrete initial velocity settings. Ultimately, the candidate coordinated strike sequence will be considered as a coordinated strike sequence. In a coordinated strike sequence, each projectile contains a defined initial launch velocity. Initial ballistic inclination angle Launch waiting time and expected hit time , as the direct command for the fire control of the electromagnetic gun.

[0097] The overall execution flow of the method of this invention is as follows: Upon receiving the coordinated strike mission, the MDN-MLP concatenated network model trained offline in step 4 is first loaded. Then, step 6-1 is executed, sequentially calling the MDN-MLP concatenated network model and the geometric correction algorithm from step 5 for each discrete initial velocity level to generate a candidate solution set. Next, the coordinated strike sequence generation algorithm from step 6-2 is executed to select the coordinated strike sequence that satisfies the launch interval constraint from the candidate solutions. Finally, the coordinated strike sequence is output. Each projectile contains a defined initial velocity, initial trajectory angle, firing delay time, and expected hit time, serving as direct commands for the electromagnetic gun's fire control.

[0098] Analysis of Result of Example: Following the above procedure Figure 3 The coordinated strike trajectory diagram for example (5000, 10000) shows the flight trajectories of three projectiles; Figure 4 The cooperative strike Gantt chart planned for the example (5000, 10000) shows the launch waiting time, reloading time and flight time of each projectile; Figure 5The coordinated strike trajectory diagram for example (20000, 3000) shows the flight trajectories of 5 projectiles; Figure 6 This is the Gantt chart for the coordinated strike planned in Example (20000, 3000). Example (5000, 10000) generates the coordinated strike trajectory of 3 projectiles as follows: Figure 3 As shown, this example has a large tilt angle, and the low initial gear prevents the projectile from reaching the target point, thus only 3 cooperative projectiles are generated. The cooperative strike scheme of the embodiment (5000, 10000) is as follows... Figure 4 As shown, the first projectile is launched at 38.10 seconds, and the shortest interval between the first and last projectiles is 3.1 seconds, which is greater than the minimum launch cycle. This example requires 11.89 seconds for a complete coordinated strike sequence. Example (20000, 3000) generates a coordinated strike scheme for 5 projectiles as follows: Figure 5 As shown, this example demonstrates the superior solution area of ​​this method. Traditional traversal methods tend to skip candidate solutions due to excessively large step sizes, and their long solution times make them unsuitable for real-time requirements. The example (20000, 3000) cooperative attack scheme is as follows: Figure 6 As shown, the first projectile is launched at 18.38 seconds, and the shortest interval between the first and last projectiles is 2.2 seconds, which is greater than the minimum launch period and makes full use of candidate solutions. The complete coordinated strike sequence requires 31.62 seconds.

[0099] The above two embodiments further verify the beneficial effects of the present invention: First, the MDN-MLP concatenated network only requires two forward propagations to obtain the initial ballistic inclination angle prediction and flight time prediction, which significantly improves the solution efficiency of the inverse ballistic problem compared with the traditional numerical traversal method and can adapt to the real-time requirements of rapid strikes; Second, the geometric correction algorithm based on direction error feedback iteratively corrects the prediction results of the MDN-MLP concatenated network, so that the final candidate solution meets the impact point accuracy constraint, ensuring engineering accuracy; Third, the cooperative strike sequence generation algorithm selects the cooperative strike sequence that meets the minimum launch period constraint from the candidate solution set in polynomial time. In the embodiments, the target points (5000, 10000) and (20000, 3000) generated cooperative strike sequences of 3 and 5 projectiles, respectively, with the minimum launch intervals between two adjacent projectiles being 3.1 s and 2.2 s, respectively, both greater than the minimum launch period of 2. Furthermore, each projectile contains a clearly defined initial velocity, initial trajectory angle, firing delay time, and expected hit time, which can be directly used as instructions for the electromagnetic gun fire control system. This demonstrates that, while ensuring engineering accuracy, this invention can rapidly generate a coordinated strike sequence that satisfies both simultaneous hit and firing interval constraints, solving the problems of uncontrolled projectiles being unable to coordinate online and the high computational cost and tendency to overlook candidate solutions in traditional numerical traversal methods.

[0100] The above-described embodiments are merely illustrative of the implementation methods of the present invention, but should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for coordinated planning of simultaneous hits of multiple projectiles in an electromagnetic railgun, characterized in that, Includes the following steps: Step 1: Divide the solution process of the electromagnetic railgun multi-projectile coordinated strike planning problem into two sub-problems, including accurately solving the initial parameters of a single-shot trajectory and generating a multi-projectile coordinated strike sequence; The initial parameters of a single-shot ballistic trajectory include the initial trajectory inclination angle of the projectile and the flight time of the projectile to reach the target point. The solution process includes multi-modal parameter prediction of the inverse ballistic problem and geometric correction of residual error. Step 2: Set the physical constraints that must be satisfied when solving for the initial ballistic parameters; Physical constraints include initial velocity discrete position constraints, initial trajectory inclination angle boundary constraints, trajectory rationality constraints, and impact point accuracy constraints. Step 3: To obtain accurate projectile flight trajectory and timing information, an external ballistic dynamics model is established based on the flight motion characteristics of the uncontrolled projectile, and a training dataset for training the MDN-MLP concatenated network is constructed accordingly; specifically: Step 3-1: Based on the characteristics of coordinated strike motion, establish an external ballistic dynamics model for the electromagnetic railgun; Step 3-2: Generate a training dataset for training the MDN-MLP concatenated network based on the external ballistic dynamics model of the electromagnetic gun. The training dataset includes MDN network training samples and MLP network training samples. Step 4: Construct an MDN-MLP concatenated network based on the MDN and MLP networks. Using the training dataset obtained in Step 3, establish a fast method for solving ballistic initial parameters based on the MDN-MLP concatenated network, resulting in the trained MDN-MLP concatenated network. Specifically: Step 4-1: Establish an MDN network to predict the initial ballistic inclination angle and obtain the predicted value of the initial ballistic inclination angle; Step 4-2: Establish an MLP network to predict the flight time of the projectile to the target point and obtain the predicted value of the flight time of the projectile to the target point; Step 4-3: Train the MDN network from Step 4-1 and the MLP network from Step 4-2 offline using the Adam optimizer, setting the initial learning rate and batch size; and save the network weights of the trained MDN and MLP networks. Step 4-4: After the training processes in Steps 4-1 and 4-2 are completed, only the network weights from Step 4-3 are loaded to construct the MDN-MLP concatenated network. Step 5: Design a geometric correction algorithm based on direction error feedback. The geometric correction algorithm receives the prediction result output from Step 4, and outputs parameters that satisfy the impact point accuracy constraint after iterative correction, thus obtaining the corrected single-shot ballistic candidate solution that satisfies the impact point accuracy constraint. The geometric correction algorithm uses the angle relationship between the deviation vector between the predicted impact point and the target point and the velocity vector at the end of the projectile, combined with the current tilt angle in the ascending or descending phase of the trajectory, to adaptively determine the correction direction and magnitude of the tilt angle and flight time. Step 6: Establish multi-missile coordinated strike sequence planning rules, and select coordinated strike sequences that satisfy the launch interval constraints from the single-shot ballistic candidate solutions in Step 5; The multi-missile coordinated strike sequence planning rule consists of two steps. First, it iterates through all discrete initial velocity levels and calls the MDN-MLP concatenated network in step 4 and the geometric correction algorithm in step 5 for each level to generate all single-shot ballistic candidate solutions that meet the impact point accuracy constraints, forming a candidate solution set for hitting the target point. Then, a greedy rule with the goal of maximizing the number of coordinated sequences is used to select sequences that meet the adjacent launch time interval constraints from the candidate solution set, which are regarded as candidate coordinated strike sequences. Finally, the initial velocity, initial trajectory inclination angle, launch waiting time and expected hit time of each projectile are output.

2. The electromagnetic railgun multi-projectile simultaneous hit cooperative planning method according to claim 1, characterized in that, In step 2, the physical constraints are specifically as follows: The initial velocity discrete gear constraint: constrains the initial velocity value to belong to the candidate initial velocity discrete gear set; let the candidate initial velocity discrete gear set be... for: (1) in, This indicates the number of gears, which is determined by the actual performance of the electromagnetic gun. For the first The initial velocity value corresponding to each gear; This is the initial velocity value corresponding to the first gear; This is the initial velocity value corresponding to the second gear; For the first The initial velocity value corresponding to each gear; The initial trajectory inclination angle boundary constraint: the initial trajectory inclination angle of the projectile's outer trajectory. The elevation angle limit of the electromagnetic railgun launching mechanism must be met, and its range is set as follows: (2) in, The initial trajectory angle of the projectile; and These are the minimum and maximum pitch angles allowed by the electromagnetic railgun launching mechanism, respectively. The ballistic rationality constraint imposes an upper limit constraint on the projectile's angle of attack and its rate of change. (3) in, For the angle of attack of the projectile; Let be the rate of change of the projectile's angle of attack; and These are the set upper limit for angle of attack and the upper limit for rate of change of angle of attack, respectively; The impact point accuracy constraint is that the Euclidean distance between the impact point and the target point corresponding to the calculated initial ballistic parameters is less than a preset threshold. ,Right now: (4) in, The landing point corresponding to the calculated initial ballistic parameters; For the target point; The preset threshold has a range of [value]. .

3. The electromagnetic railgun multi-projectile simultaneous hit cooperative planning method according to claim 2, characterized in that, Step 3-1 specifically involves: Assuming the unguided projectile fired by the electromagnetic railgun has no lateral maneuvering during its flight; for a given incoming direction angle of... For a single target, all projectiles fire from the same direction at an angle of 0°. Flight within the vertical plane; a three-degree-of-freedom rigid body ballistic model in the vertical plane is adopted, which discards the lateral motion and lateral moment-related components while fully preserving the dynamic characteristics of the projectile's pitch attitude; A ground coordinate system is established with the initial position of the center of mass at the projectile launch point as the origin. After dimensionality reduction, the projectile only undergoes translational motion of its center of mass and pitch rotation around the horizontal axis in the vertical plane. The external ballistic dynamics model of the electromagnetic railgun is as follows: (5) in, The coordinates of the projectile's center of mass in the ground coordinate system within the impact plane; For the projectile's flight time; The velocity of the projectile's center of mass; , These are the trajectory inclination angle and the elevation angle, respectively. It is the pitch angular velocity; For the mass of the projectile; For pitch rotation inertia; , , These are the drag, lift components, and pitching moment of aerodynamics in the ballistic coordinate system; specifically: (6) in, For the angle of attack; air density; The reference area of ​​the projectile; The characteristic length of the projectile; This is the drag coefficient; This is the derivative of the lift coefficient with respect to the angle of attack; and These are the derivatives of the pitch moment coefficient with respect to the angle of attack and with respect to the dimensionless pitch velocity, respectively. The initial conditions for the external ballistic dynamics model of the electromagnetic railgun are: , , , .

4. The electromagnetic railgun multi-projectile simultaneous hit cooperative planning method according to claim 3, characterized in that, Step 3-2 specifically involves: Step 3-2-1: Set of candidate initial velocity discrete gears set in step 2 and initial trajectory inclination boundary constraints Uniform random sampling is performed within the defined parameter space to obtain sampling parameters, resulting in several sets of initial parameter combinations. ,in For the first The initial velocity values ​​corresponding to each gear. For the first The initial trajectory inclination angle of each sample is obtained; for each set of sampled parameters, the fourth-order Runge-Kutta method is used to solve the electromagnetic gun external ballistic dynamics model in step 3-1 to obtain numerical simulation data, obtain the impact point coordinates and projectile flight time corresponding to the impact point, and uniformly sample the trajectory within its flight time range to obtain the sampled dataset. Step 3-2-2: Based on the sampling dataset obtained from numerical simulation data, construct two types of training samples, including MDN network training samples and MLP network training samples; The process of determining the training samples for the MDN network is as follows: the input is the coordinates of the target point. With initial velocity The output is the true value of the initial ballistic inclination angle obtained from sampling. That is, to establish a mapping relationship ;in, Indicates the first The coordinates of the landing points of the samples are used to obtain the training samples for the MDN network. MDN network training samples are used to train the MDN network so that it can infer the initial trajectory inclination angle corresponding to the ballistic trajectory based on the target position and initial velocity. The process of determining training samples for an MLP network is as follows: the input is the coordinates of the target point. initial velocity True value of the initial trajectory inclination angle The output is the destination point. Flight time is truly valuable That is, to establish a mapping relationship Finally, the training samples for the MLP network were obtained. MLP network training samples are used to train the MLP network so that it can predict the time required for a projectile to reach the target based on the target position, initial velocity, and tilt angle. Step 3-2-3: Normalize both the MDN network training samples and the MLP network training samples to obtain the normalized MDN network training samples and MLP network training samples, which are the training datasets used for training the MDN-MLP concatenated network.

5. The electromagnetic railgun multi-projectile simultaneous hit cooperative planning method according to claim 4, characterized in that, In step 3-2-1, the sampled dataset is resampled using the uniform density box principle, so that the sampled dataset is within the uniform density box. It is approximately uniformly distributed within the region that can be represented.

6. The electromagnetic railgun multi-projectile simultaneous hit cooperative planning method according to claim 5, characterized in that, Step 4 specifically involves: Step 4-1 specifically involves: Hybrid density network, or MDN network for short, uses three neurons as the input layer to receive the first three values ​​from the normalized MDN network training samples in step 3-2. ,use Each layer is a fully connected hidden layer. Each neuron uses ReLU nonlinear activation between layers, outputting a mixing coefficient. mean With log standard deviation The parameters of the Gaussian mixture model; where the number of Gaussian components in the MDN network. The value is set to 2, corresponding to the high-angle and low-angle ballistic trajectories respectively; the MDN network uses the negative log-likelihood loss function (NLL) to fit the multimodal probability distribution. The expression is: (7) in, For training batch size, For batch indexing; Index of Gaussian components; For the first The sample at the th The mixing coefficient over each Gaussian component; and The first The sample at the th Mean and standard deviation of each Gaussian component; For the first The ground truth value of the initial trajectory inclination angle for each sample is the same as the ground truth value of the initial trajectory inclination angle in the training dataset. ; It is the probability density function of the standard normal distribution; After training the MDN network using training samples, the mean value of the low-extension trajectory is selected when using the MDN network for prediction. The output is used as the initial trajectory inclination angle prediction value for this trajectory. ; Step 4-2 specifically involves: A fully connected network, or MLP network for short, uses four neurons as its input layer, receiving the first four values ​​from the normalized MLP network training samples in step 3-2. The hidden layer is layer The nodes are fully connected layers, with ReLU nonlinear activation between layers. The network output layer outputs the predicted flight time of the projectile to the target point. During training, the output layer will receive data from the training samples of the MLP network. MLP networks are trained using the Mean Squared Error (MSE) loss function. The expression is: (8) in, For MLP networks to the first The predicted flight time of a projectile to the target point for a sample; For the first The ground truth value of the projectile's time to reach the target point for each sample, compared with the ground truth value of the predicted time to reach the target point for each projectile in the training dataset. ; Step 4-4 specifically involves: When using it online, first enter through the MDN network. Output at Initial velocity reached Predicted initial trajectory inclination angle for low-profile ballistics Then input through the MLP network Solve for the initial trajectory inclination angle The corresponding predicted flight time to the target point The predicted initial trajectory inclination angle and the predicted flight time of the projectile to the target point constitute the trajectory parameters, which serve as the prediction results.

7. The electromagnetic railgun multi-projectile simultaneous hit cooperative planning method according to claim 6, characterized in that, Step 5 specifically involves: Step 5-1: Initialize the parameters of the geometric correction algorithm based on direction error feedback, and calculate the deviation vector and direction difference angle; make Set the maximum number of iterations. ; Assuming that when using an MDN-MLP concatenated network, the input is the real target point. With a given initial velocity The ballistic parameters to be corrected are obtained as the predicted values ​​of the initial ballistic inclination angle. Compared with flight time prediction ; Using the electromagnetic railgun external ballistic dynamics model from step 3-1, integrate to At that moment, the predicted landing point was obtained. And the projectile's terminal velocity vector at that moment is ;in These represent the horizontal distance and height of the actual target point, respectively. These represent the horizontal distance and height of the predicted landing point, respectively. These are the horizontal and vertical components of the terminal velocity, respectively. Define from the predicted landing point point to deviation vector for: (9) Current landing point error for: (10) in, , They are respectively directional deviation component; Calculate the deviation vector With the terminal velocity vector Direction difference angle for: (11) in, The four-quadrant arctangent function returns a value in degrees. Will Standardize the quadrant transformation operation. Within the interval; if Corresponding landing point error Less than the preset threshold error If the correction is skipped, the initial parameters of the single-shot trajectory predicted by the MDN-MLP concatenated network will be directly applied. Output directly as a candidate solution for single-shot trajectory; otherwise, correction is required, proceed to step 5-2; the initial parameters for the single-shot trajectory in the current 0th iteration are... ; Step 5-2: When performing Step 5-1 to determine if correction is needed, determine whether the direction of angle correction needs to be reversed based on the vertical component of the initial trajectory inclination angle and the terminal velocity; for the different characteristics of high-angle and low-angle trajectories, a large-angle reversal rule is introduced; specifically: Based on the initial ballistic inclination angle prediction value obtained in step 5-1 The vertical component of the current terminal velocity is Determine the angle sign factor; define the angle sign factor. In order to satisfy and Time to take In other cases, take To achieve a global reversal of the angle correction direction, where A dimensionless symbolic factor; Step 5-3: Confirm after performing step 5-2. Then, determine the angle correction sign and time correction sign based on the quadrant where the direction difference angle is located, and calculate the correction magnitude, including the angle correction amount and the time correction amount; according to Determine the angle correction sign within the specified interval. and time correction symbols Both are dimensionless symbolic factors; Define the landing point sensitivity coefficient , No. Angle correction amount in the next iteration and time correction amount They are respectively: (12) (13) in, For the first Error at the landing point of the next iteration; For the first Next iteration direction difference angle; For the first The next iteration predicts the height of the landing point; For the first The horizontal component of the velocity at the end of the next iteration; For the first Vertical component of velocity at the end of the next iteration; Step 5-4: After determining the correction magnitude for each iteration in Step 5-3, update the initial parameters of the single-shot trajectory and treat them as candidate parameters, saving the candidate parameters that satisfy the impact point accuracy constraint; calculate the candidate parameters as follows: (14) in, These are the updated initial trajectory inclination and flight time to the target point, respectively. For the first The flight time of the projectile to the target point is obtained from the next iteration; Will Forced constraints to physical candidate domains , For candidate parameters Recalculate the ballistic integral and recalculate the new error. ;like Then accept candidate parameters. ;like Reject candidate parameters Continue iterating; Step 5-5: Continue the iterations from Steps 5-2 to 5-4. If the landing point error satisfies... The correction is then deemed successful, and the candidate parameters are... Considered as a candidate solution for single-shot ballistics If the accuracy requirement is not met even after reaching the maximum number of iterations, then the candidate parameters are discarded. These are marked as unusable candidate solutions; finally, a geometric correction algorithm is used to obtain the corrected single-shot trajectory candidate solutions that satisfy the impact point accuracy constraint.

8. The electromagnetic railgun multi-projectile simultaneous hit cooperative planning method according to claim 7, characterized in that, In step 5: In step 5-1, a preset threshold error is defined. The range is ; In step 5-3: when At that time, take ;when At that time, take ;when At that time, take ;when At that time, take ; Landing point sensitivity coefficient ,in is the base of the natural logarithm; For the first The initial trajectory inclination angle obtained in the next iteration.

9. The electromagnetic railgun multi-projectile simultaneous hit cooperative planning method according to claim 8, characterized in that, Step 6 specifically involves: Step 6-1: Generate a candidate solution set for hitting the target point; For a given target point and expected hit time Traverse the discrete gear set of initial velocity ; For each discrete initial velocity level The target point coordinates are input through the trained MDN-MLP concatenated network in step 4. and initial velocity The initial ballistic inclination angle prediction and the projectile's flight time to reach the target point are obtained. ; The geometric correction algorithm based on direction error feedback from step 5 is used for iterative correction. If the landing point error is corrected... This yields a candidate solution for a single-shot ballistic trajectory. ,in For the first A corrected initial trajectory inclination angle For the first The corrected flight time of the projectile to reach the target point; Calculate the first Candidate solution launch waiting time That is, the time required from the current moment until the launch of the projectile, where, For the expected hit time; if Then the candidate solutions for single-shot ballistics will be retained; After traversing all discrete initial velocity levels, all retained single-shot trajectory candidate solutions constitute the target hit point candidate solution set. ,in This represents the number of candidate solutions for a single-shot ballistic trajectory. Indicates the first A single-shot ballistic candidate solution This is the first candidate solution for a single-shot trajectory. This is the second candidate solution for a single-shot trajectory. For the first One single-shot ballistic candidate solution; Step 6-2: Construct a cooperative strike sequence generation algorithm based on the candidate solution set of hit target points obtained in Step 6-1; The candidate solution set of hit target points obtained in step 6-1 All candidate solutions are ranked according to flight time. Arranged in ascending order, with shorter flight times listed first; assuming the minimum firing period of the electromagnetic railgun is... , The minimum time interval required between the launch times of two adjacent projectiles is determined by the physical characteristics of the artillery loading and resetting mechanism. A greedy rule with the goal of maximizing the number of cooperative sequences is adopted to select candidate cooperative strike sequences that meet the launch interval constraints from the candidate solution set of target hit points. That is, a cooperative strike sequence generation algorithm is designed to obtain the cooperative strike sequence.

10. A method for coordinated planning of simultaneous hits of multiple projectiles in an electromagnetic railgun according to claim 9, characterized in that, In step 6-2, the cooperative strike sequence generation algorithm is specifically as follows: Initialize the coordinated strike sequence Given an empty sequence, iterate through the sorted candidate solution set for hitting the target point. In ,like If empty, then Join the coordinated strike sequence ;like If the value is not empty, it proves that there exists a single-shot ballistic candidate solution in the sequence. The last single-shot ballistic candidate solution added in the middle is , The corresponding launch waiting time is Then the first Launch latency of a single-shot ballistic candidate solution ;in, For the last time to join the coordinated strike sequence The initial velocity of the candidate solution for the single-shot trajectory. For the last time to join the coordinated strike sequence The initial trajectory inclination angle after correction of the candidate solution for a single-shot ballistic trajectory. For the last time to join the coordinated strike sequence The flight time of a single-shot ballistic candidate solution after correction to the projectile's trajectory to the target point; If satisfied Then the current single-shot ballistic candidate solution will be... Join the coordinated strike sequence and update Otherwise, skip the current single-shot trajectory candidate solution and continue iterating through subsequent single-shot trajectory candidate solutions; where, For the last time joined The launch latency of a single-shot ballistic candidate solution; This is the minimum firing cycle for the electromagnetic railgun; After the traversal is complete, the coordinated strike sequence That is, a candidate cooperative strike sequence that satisfies both the simultaneous hit and launch interval constraints; if If empty, it indicates that a coordinated strike plan cannot be planned under the current initial velocity dispersion level and range limitations, and adjustments are needed. Alternatively, consider expanding the initial velocity discrete range configuration; ultimately, consider candidate coordinated strike sequences as coordinated strike sequences; in a coordinated strike sequence, each projectile contains a defined initial launch velocity. Initial ballistic inclination angle Launch waiting time and expected hit time , as the direct command for the fire control of the electromagnetic gun.

Citation Information

Patent Citations

  • Multi-aircraft time-sharing attack cooperative guidance method

    CN116880560B

  • Machine learning based vehicle dynamics parameter estimation method

    CN119647286B