Method for optimizing cooperative path planning of flight weapon cluster based on guidance solution generation group

Abstract

The present application relates to path planning technical field, especially based on guide solution generation group optimization flight weapon cluster cooperative path planning method, including setting flight weapon cluster cooperative path planning scene;Based on the original path of the terrain model and the equivalent mountain terrain model constructs the mathematical model of map fusion;Flight weapon cluster path parameterization, determine optimization target and constraint condition;The geometric characteristics of the explicit mining Pareto solution set are approached by using self-organizing mapping manifold, and the neuron weight distribution is constructed by incremental weight updating, the individual optimal and global optimal guide solution are predicted, the path population is updated by combining adaptive particle swarm optimization iteration, and the flight weapon cluster cooperative path planning is outputted.The present application solves the problem that the existing multi-objective particle swarm optimization method is limited in constructing and utilizing effective guide information in the group evolution process, is easily affected by the population distribution fluctuation, thereby restricting the convergence efficiency and solution set quality of the algorithm.

Description

Technical Field

[0001] This invention relates to the field of path planning technology, and in particular to a collaborative path planning method for flight weapon clusters based on guided solution generation and swarm optimization. Background Technology

[0002] In modern warfare, airborne weapon clusters need to coordinate their missions in complex and ever-changing battlefield environments to effectively strike targets and reduce the risk of interception and mutual interference. To this end, airborne weapon clusters typically need to conduct coordinated path planning to ensure that each member reaches the target area according to the predetermined time sequence and spatial relationships, while meeting flight safety and operational constraints. Therefore, how to achieve efficient and reliable coordinated path planning for airborne weapon clusters under complex constraints has become a key technical problem in the field of path planning.

[0003] Unlike traditional single-vehicle trajectory planning, collaborative path planning for flight weapon clusters not only needs to focus on the individual optimality of each flight weapon, but also needs to meet the constraints of time synchronization, spatial distribution coordination, and task collaboration among members at the global level. This makes the path planning problem exhibit high-dimensionality, multiple constraints, and multiple objectives, which places higher demands on the global search capability and stability of the planning method.

[0004] To address the aforementioned issues, particle swarm optimization (PSO) and its multi-objective extensions are widely used in optimization problems such as path planning due to their superior parallel search capabilities. However, under complex constraints and battlefield environments, existing multi-objective PSO methods have limited ability to construct and utilize effective guidance information during population evolution and are easily affected by population distribution fluctuations, thus limiting the algorithm's convergence efficiency and solution quality. Summary of the Invention

[0005] To address the shortcomings of existing methods, this invention solves the problem that existing multi-objective particle swarm optimization methods have limited ability to construct and utilize effective guidance information during population evolution, and are easily affected by population distribution fluctuations, thus restricting the convergence efficiency and solution set quality of the algorithm.

[0006] The technical solution adopted in this invention is: a collaborative path planning method for flight weapon clusters based on guided solution generation and swarm optimization, comprising the following steps: Step 1: Set up a scenario for collaborative flight path planning of flight weapon clusters; Step 2: Construct a mathematical model for map fusion based on the original flight path terrain model and the equivalent mountain peak terrain model; As a preferred embodiment of the present invention, the mathematical model for map fusion is formulated as follows: in, Point The corresponding mountain height, Point The corresponding terrain elevation.

[0007] In a preferred embodiment of the present invention, the formula for the height of the mountain peak is: in, Represents the x and y coordinates of a point on a horizontal plane; Indicates the number of mountain peaks; Indicates the first The height of the mountain peak; Indicates the first The horizontal and vertical coordinates of the center of the mountain peak on the horizontal plane; , Indicates the first The contour parameters of the mountain peak.

[0008] Step 3: Perform flight path parameterization for the flight weapon cluster and determine the optimization objectives and constraints; Step 4: Explicitly mine the geometric properties of the Pareto solution set by approximating the self-organizing map manifold, and construct the neuron weight distribution through incremental weight update to predict the individual optimal and global optimal guided solution. Combined with adaptive particle swarm optimization, iteratively update the path population and output the path planning that satisfies the collaborative path planning of the flight weapon cluster. In a preferred embodiment of the present invention, step four specifically includes: Step 41: Initialize the particle swarm; Step 42: Construct an incremental SOM model. In each iteration, the current path population is used as the training input. BMU search, neighborhood function calculation, and single-step weight update are performed sequentially on each particle in the population. The neighborhood radius and latent space dimension are set. The first The initial weights of the SOM are defined as the weight states of generation t-1; For any path sample in the current generation path population Find its BMU ; Calculate the neighborhood influence of nodes surrounding BMU ; Define a path comprehensive evaluation function; In a preferred embodiment of the present invention, the formula for the path comprehensive evaluation function is as follows: in, For path The average or cumulative flight elevation along the path; , These are the normalization coefficients; These are the weighting coefficients.

[0009] Construct a weight update modulation function based on the path comprehensive evaluation function; In a preferred embodiment of the present invention, the formula for the weight update modulation function is: in, This is the modulation intensity parameter.

[0010] Perform a weight update; In a preferred embodiment of the present invention, the formula for weight update is: in, It is the first Learning rate.

[0011] Step 43: Map the particles in the current population to the model latent space, and predict and generate a guiding solution based on the mapping results to guide the multi-objective particle swarm search. In a preferred embodiment of the present invention, step 43 specifically includes: Will The weight vector of each neighboring neuron in the decision space The individual optimal guided solution is generated by weighting the average of the reciprocals of the magnitudes of the weight vectors. The formula is: in, These are the weighting coefficients.

[0012] In a preferred embodiment of the present invention, the formula for the weighting coefficient is: In the formula, The second norm of a vector; It is a preset minimum positive number.

[0013] gbest guides the generation of solutions.

[0014] Step 44: Update the particle's velocity and position using the guided solution; Step 45: Update the external archive; Step 46: Stop the iteration when the preset shutdown conditions are met.

[0015] As a preferred embodiment of the present invention, a guided solution generation swarm optimization collaborative path planning system for flight weapon clusters includes: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement a guided solution generation swarm optimization collaborative path planning method for flight weapon clusters.

[0016] The beneficial effects of this invention are: 1. This invention models the collaborative path planning problem of flight weapon clusters as a multi-objective optimization problem. By constructing a path planning model that includes multi-objective evaluation indicators such as track length and flight altitude, it achieves comprehensive optimization of collaborative tracks and provides feasible and effective collaborative path planning solutions for flight weapon clusters. 2. In the process of particle swarm evolution, an incremental self-organizing mapping model is introduced to learn the distribution characteristics of the solution space generation by generation, and predicts the individual optimal guiding solution and the global optimal guiding solution based on the best matching unit and its neighborhood information, thereby providing an effective search direction for the particle swarm search process, guiding particles to evolve towards the potential optimal solution region, and improving the swarm search efficiency. 3. By adopting an incremental learning approach to update the self-organizing map model online, the model is completely retrained in each iteration, which reduces computational complexity and improves computational efficiency while ensuring the accuracy of collaborative path planning. 4. The method of the present invention is applicable to cluster collaborative path planning scenarios of various flight platforms such as UAV swarms and cruise missile formations, and has good versatility and engineering application value; 5. Construct a multi-peak exponential function to equivalently model battlefield threats, and uniformly map different types of threats such as mountain peaks and enemy defense zones into a continuous height field, thus realizing a unified mathematical expression of complex threat environments. Attached Figure Description

[0017] Figure 1 This is a flowchart of the general scenario flight weapon cluster collaborative path planning method based on guided solution generation and group optimization of the present invention; Figure 2 This is a schematic diagram of the multi-objective particle swarm guided solution prediction process based on incremental SOM; Figure 3 It is the final population approximation of the Pareto front obtained by the IPPSO algorithm (this invention) when independently running the index value on the ZCAT test problem; Figure 4 This is the flight path map planned by this invention; Figure 5 This is a top view of the flight path map planned in this invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments. The drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0019] like Figure 1 As shown, the guided solution generation-based swarm optimization method for collaborative path planning of flight weapon clusters includes the following steps: First, in terms of modeling, based on the pre-set combat scenario, the combat area, starting and target positions, terrain environment and flight constraints of the flight weapon cluster are uniformly described. The collaborative path planning problem of the flight weapon cluster is modeled as a multi-objective optimization model containing multiple optimization objectives and collaborative constraints, so that the complex collaborative path planning problem can be represented and solved in a unified mathematical form.

[0020] Secondly, in terms of method design, addressing the problem that most existing multi-objective particle swarm optimization algorithms and their improvements are based on the "selection" paradigm to determine individual optimal solutions and global optimal solutions, which easily leads to premature convergence when the population lacks high-quality solutions, this invention proposes a guidance mechanism based on the "prediction" paradigm. During the swarm optimization iteration process, a self-organizing map model is introduced to learn the distribution characteristics of the current population solution set, and the low-dimensional manifold structure of the Pareto solution set in the multi-objective optimization problem is gradually approximated through incremental updates. On this basis, based on the learned solution space topology, individual optimal guiding solutions and global optimal guiding solutions are predicted and generated to guide the evolution of the particle swarm, and the guiding solutions are introduced into the particle position and velocity update process to guide the particle swarm to search for potential excellent solution regions.

[0021] By integrating the construction of a collaborative path planning model with a guided solution prediction mechanism based on incremental self-organizing mapping, this invention can provide an efficient and stable optimization solution for collaborative path planning applications of flight weapon clusters under complex combat conditions. The cooperative path planning problem for airborne weapon clusters is modeled as a multi-objective optimization problem, where optimization objectives may include trajectory length, flight altitude, etc. The method utilizes the swarm cooperation characteristic of particle swarm optimization to search the optimal solution space, with each particle representing a candidate path scheme. The search direction is guided by predicting the individual best (pbest) and global best (gbest) solutions of each particle, thereby improving the algorithm's convergence and diversity. The algorithm approximates the manifold structure of the Pareto solution set in a low-dimensional latent space using a self-organizing map (SOM), and uses the learned structure to predict the guided solution, mining and utilizing the regularity of the Pareto solution set in multi-objective optimization to improve optimization performance. This method combines guided solution prediction with SOM manifold modeling, providing a new solution approach for cooperative path planning of airborne weapon clusters, and has strong application value.

[0022] Specifically, to address the slow convergence and insufficient diversity issues caused by the reliance on random sampling of the target space in traditional MOPSO, this invention introduces a self-organizing map to explicitly approximate the (m-1)-dimensional manifold structure of the Pareto solution set. Unlike the complete batch training of traditional SOM, this invention employs an incremental training mechanism: after each generation update of the particle swarm, a "BMU search—neighborhood function calculation—single-step weight update" is performed with the current path population as input, updating the latent space neuron weights generation by generation. This incremental update method reduces additional computational overhead while maintaining topological stability, enabling SOM to continuously approximate the Pareto manifold and maintain sensitivity to population distribution. SOM maps the path population in the high-dimensional decision space to the low-dimensional latent space and continuously adjusts the neuron weights through incremental updates to capture the overall distribution pattern.

[0023] Building upon this, individual optimality (pbest) is generated by mapping particles to the latent space and reading their corresponding neuron weight vectors, achieving a direct correlation with the solution space geometry; global optimality (gbest) is obtained by randomly sampling weight vectors from neighboring neurons in the latent space, taking into account both convergence trend and population diversity. The guided solution generation mechanism based on incremental SOM enables particles to search along the manifold structure direction, thereby improving the efficiency and stability of multi-objective optimization.

[0024] See appendix Figure 1 This invention treats the cooperative path planning problem of a flight weapon cluster as a multi-objective optimization problem. It explicitly mines the geometric characteristics of the Pareto solution set by approximating the self-organizing map (SOM) manifold and constructs the neuron weight distribution by incremental weight update to predict the individual optimal (pbest) and global optimal (gbest) guided solution. Combined with adaptive particle swarm optimization, it iteratively updates the path population and finally outputs a cooperative path planning scheme that satisfies the flight weapon cluster.

[0025] Step 1: Setting up operational scenarios: Based on the operational scenarios, set up a flight weapon cluster collaborative trajectory planning scenario, and clarify the operational area, starting point and target point positions and related flight constraints; Construct a horizontal range of The height range is A three-dimensional combat zone is defined; within this combat zone, a flight weapon cluster consisting of 5 drones is set up. Each drone takes off from one side of the combat zone and must fly towards the predetermined target area at a flight altitude of not less than 0.5 km and not more than 4 km.

[0026] Step 2: Equivalent digital modeling of environment and terrain: The original terrain information and threat elements in the combat scenario are represented by equivalent digital representation to construct a three-dimensional digital battlefield environment that can be used for path planning calculations, so as to support the modeling and solving of flight weapon cluster trajectory planning problems. The digital terrain that simulates the original flight path is calculated using the following formula: (1) In the formula, Represents the x and y coordinates of a point on a horizontal plane; Point Corresponding terrain elevation; five constants This represents the terrain coefficient; by adjusting the terrain coefficient, various different terrains and landforms can be simulated, and the simulated terrains and landforms can be used as the known terrains and landforms for flight path planning. Aerial weapons pass through this The area contains 5 threats; the original terrain is calculated according to Formula 1: The original digital terrain was obtained through calculation, as follows: Figure 4 As shown.

[0027] Cruise missiles typically encounter various threats during flight, such as mountains and enemy defense zones. For the convenience of aviation regulations, this invention incorporates all threats into the planning model as equivalent mountain terrain. The mountain terrain is generated by the following formula: (2) in, Represents the x and y coordinates of a point on a horizontal plane; Point The corresponding mountain peak height; Indicates the number of mountain peaks; Indicates the first The height of the mountain peak; Indicates the first The horizontal and vertical coordinates of the center of the mountain peak on the horizontal plane; , Indicates the first The contour parameters of the mountain peaks; by changing the parameters in the formula, various numbers and shapes of threatening equivalent terrain can be simulated; This invention uses a multi-peak exponential function to equivalently model battlefield threats, mapping different types of threats, such as mountains and enemy defense zones, into a continuous altitude field, thus achieving a unified mathematical expression for complex threat environments. This model is continuous and smooth, avoiding the search instability problems caused by discrete no-fly zones or hard constraints, and is suitable for combination with continuous optimization algorithms such as multi-target particle swarm optimization. By adjusting the mountain height, center position, and contour parameters, it can flexibly simulate threat areas of different quantities, intensities, and spatial distributions, improving the versatility and engineering applicability of the trajectory planning model while ensuring computational efficiency.

[0028] Utilizing the principle of digital map information fusion, a threat-equivalent map and the original digital terrain are fused to generate an equivalent digital map. The mathematical model for map fusion is as follows: (3) Step 3: Parameterize the trajectory and determine the optimization objectives and constraints; In the digital battlefield environment, the optimization objectives, constraints, and optimization variables for the collaborative flight path planning of flight weapon clusters are determined. The optimization objectives include at least the total flight path length and flight altitude, the optimization variables are the spatial coordinate parameters of the flight path points, and the constraints include the maximum flight altitude, the minimum flight altitude, the minimum flight path segment length, and the maximum total flight path length. The variable that needs optimization is all waypoints (excluding the starting and ending waypoints). coordinate; Calculate two objective function values ​​for each particle: the shorter the total trajectory length, the better; and the lower the flight altitude, the better. All particles are used to represent a complete flight path of a flight weapon, and the position information of each particle is composed of the three-dimensional coordinates of several intermediate path points. Together they constitute the whole; the waypoints participate as optimization variables in the multi-objective particle swarm optimization process; the particle swarm algorithm achieves the search and optimization of the complete waypoint scheme by updating the particle position vectors as a whole. make Represents the waypoint; then These represent the starting point and target point of the flight weapon, respectively; ,···, express A waypoint obtained through planning Indicates the first The three-dimensional coordinates of each waypoint; therefore, the total length of the waypoint is expressed as: (4) in, Indicates the first and The distance between each waypoint.

[0029] The optimization objective for flight altitude is expressed as: (5) In the formula, For the first Flight altitude at each waypoint.

[0030] Based on the digital environment constructed in step two, the UAV flight path is parameterized. In this embodiment, the flight path of each UAV is discretized into 10 track points (including start and end points), and adjacent track points are connected sequentially to form a complete flight path. Each track point is represented by three-dimensional coordinate parameters. This means that a single drone corresponds to 30 track parameters, and 5 drones constitute 150 track parameters.

[0031] Based on the parameterization of the flight paths, the optimization objectives and constraints of the collaborative path planning are determined. Among them, the sum of the flight path lengths of the five UAVs is used as the evaluation index of the total flight path length of the cluster; and the average height of all flight path points is used as the evaluation index of the average flight altitude of the cluster. The altitude of the waypoints is always kept within the range of 0.5 km to 4 km; the spatial distance between adjacent waypoints is not less than 5 km; and the total length of the track of a single UAV is not more than 250 km.

[0032] The aforementioned objectives and constraints will serve as a unified evaluation and judgment basis in subsequent optimization processes; Step 4: Construct a multi-objective optimization model: Based on the optimization objectives, constraints, and optimization variables, the collaborative trajectory planning problem of the flight weapon cluster is uniformly modeled as a multi-objective optimization problem, forming a complete collaborative path planning optimization model for the flight weapon cluster. The multi-objective optimization model takes a 150-dimensional trajectory parameter vector as the input of optimization variables and outputs the evaluation results of the total length of the cluster trajectory and the average flight altitude of the corresponding cooperative trajectory scheme, which are used for subsequent optimization solutions. Step 5: Initialize the particle swarm optimization model by randomly generating particles containing... An initial population of particles, where each particle represents a feasible trajectory for a flight weapon; Randomly generated The group cooperative track scheme is used as the initial population; during the generation process, the following constrained random sampling is performed on each track point: track points coordinates in the interval Uniformly randomized generation within the track points; coordinates in the interval The path is randomly generated; after generation, the distance between adjacent path points is checked. If the length of a path segment is less than... If so, the corresponding waypoint position will be readjusted.

[0033] Step 6: Construct an incremental SOM model: In each iteration, the self-organizing map method of incremental learning is adopted. While retaining the parameters of the previous generation model, the model is updated online only once based on the current generation particle samples, so as to continuously learn the distribution characteristics of the solution space and approximate the manifold structure of the Pareto solution set in the multi-objective optimization problem. In step six, the SOM incremental learning mechanism is adopted: in each iteration, the current path population (including...) is... N Using 100 particles as training input, incremental SOM training is completed by sequentially performing "BMU search - neighborhood function calculation - single-step weight update" on each particle in the population, without having to perform multiple complete iterations in each generation. Unlike the traditional method that performs a complete SOM in each generation of the algorithm, the method of this invention uses the new population generated by the algorithm in each generation and the SOM for incremental training, so as to reduce the additional computational overhead caused by SOM manifold learning. Training parameters include: the neighborhood radius σ decays exponentially with each iteration to control the neighborhood range; the latent space dimension is dynamically configured according to the number of targets: for bi-objective problems, a one-dimensional topological structure can be used for configuration. No. The initial weights of the generation SOM are defined as the weight state at the end of the previous generation's training: (6) in, Indicates the first The middle generation After the first sample is updated, the th... The weight vector of each neuron; For any path sample in the current generation path population Find its BMU (Best Matching Unit): (7) in, The best-matching unit (BMU) is used. The current update status is the The weight vector of each neuron; Calculate the neighborhood influence of nodes surrounding BMU: (8) in, and The coordinates of the neuron. Let be the radius of the neighborhood function; To enhance the self-organizing map model's ability to perceive high-quality path samples, a joint modulation function based on path length and flight elevation is introduced. First, a comprehensive path evaluation function is defined: (9) in, For path The average or cumulative flight elevation along the path; , These are the normalization coefficients; These are the weighting coefficients; Based on the path evaluation function, a weight update modulation function is constructed: (10) in, The modulation intensity parameter; Use the following formula to perform a weight update: (11) in, It is the first The learning rate, after completing the learning rate After the sequential scan, the first... The SOM training update is now performed in a new generation, and repeated iterations are no longer required.

[0034] The 100 particles generated in step 5 are used as training samples and input into the self-organizing map model to perform an incremental online training on the SOM network. During the training process, the existing neuron weights are not reset, but the weights are updated step by step based on the current particle samples to learn the distribution characteristics of the current cooperative trajectory in the 150-dimensional solution space.

[0035] Step 7, Guided Solution Prediction: Based on the updated self-organizing map model in Step 6, the particles in the current population are mapped to the model latent space, and guided solutions are predicted and generated based on the mapping results to guide the multi-objective particle swarm search. After completing the first After training the incremental self-organizing map model, for the current path sample The corresponding Best Matching Unit (BMU) From its set of neighboring nodes Selected from We construct individual optimal guided solutions pbest and global optimal guided solutions gbest for particle swarm updating using a number of nearest neighbor neurons. Neighborhood set The BMU is composed of neighboring neurons in the latent space topology, from which the neurons with the smallest topological distance from the BMU are selected. Several neurons are involved in guiding solution generation, among which... It is an integer greater than 1.

[0036] pbest-guided solution generation: This will generate the selected solution. The weight vector of each neighboring neuron in the decision space The individual optimal guided solution is generated by weighting the average of the reciprocals of the magnitudes of their weight vectors. The calculation formula is as follows: (12) Among them, the weighting coefficient Defined as: (13) In the formula, The second norm of a vector; It is a preset very small positive number used to avoid division by zero errors.

[0037] gbest guided solution generation: The weight vector of the corresponding neuron in the BMU in the decision space is directly used as the globally optimal guided solution. (14) in, This represents the weight vector of the BMU in the decision space (i.e., the weight representation of the latent space neurons in the decision space).

[0038] In self-organizing map models, the magnitude of a neuron's weight vector in the decision space reflects the overall scale characteristics of its corresponding path solution. Weight vectors with smaller magnitudes typically correspond to regions of high-quality path solutions with shorter travel distances and higher constraint satisfaction.

[0039] By using the reciprocal of the magnitude of the weight vector as the weighting coefficient, the guiding role of neurons corresponding to high-quality paths can be adaptively enhanced when generating the pbest guided solution, making individual particles more inclined to search along low-range, low-cost paths. At the same time, by directly using the BMU weight vector as the gbest guided solution, the stability of the global search direction is maintained.

[0040] Furthermore, by dynamically adjusting the number of neighboring neurons involved in the weighting... In the early stages of the algorithm, more neighborhood information is introduced to enhance global exploration capabilities, while the amount of neighborhood information is gradually reduced in the later stages. This enhances the local development characteristics, thereby achieving an effective balance between global exploration and local convergence without increasing the complexity of particle swarm updates.

[0041] After completing the incremental training of SOM in step six, the current particle swarm is mapped to the SOM latent space. For each particle, the weight vector of its corresponding best matching neuron is selected as the individual guiding solution for that particle. At the same time, multiple neighboring neurons are selected from the neighborhood of that neuron, and their weight vectors are weighted and fused to generate a global guiding solution, which is used to guide the particle search direction.

[0042] Step 8, Particle Update: Based on the guided solution obtained in Step 7, and combined with the multi-objective particle swarm optimization algorithm, the position and velocity of the particles are updated to generate a new offspring track population, thereby realizing a guided search of the solution space. In step eight, the particle update process includes: 1. Speed ​​update formula: (15) in, Represents element-wise multiplication of vectors (Hadamard product). It is the first The coefficient of inertia over time; , These are the cognitive and social acceleration coefficients, respectively. , Each dimension is generated independently and follows a uniform distribution. A random vector; 2. Position update formula: (16) In a specific implementation of this invention, the time step Set to 1.

[0043] Feasibility repair mechanism: Geometric pruning or resampling of paths that exceed constraints to ensure that paths meet battlefield environment requirements.

[0044] Under the combined effect of individual and global guided solutions, the velocity and position of particles are updated to obtain a new cooperative trajectory scheme. After the update is completed, the following repair processes are performed on the generated trajectories in sequence: the height of trajectory points exceeding the range of [0.5 km, 4 km] is pruned; trajectory segments with a distance of less than 5 km between adjacent trajectory points are resampled; and particles with a total trajectory length of more than 250 km for a single UAV are penalized or removed.

[0045] Step 9, External Archive Update: Based on the multi-objective optimization model, evaluate and screen the newly generated child tracks, and update the current population so that the collaborative tracks of the flight weapon cluster gradually approach the Pareto optimal solution set of the multi-objective optimization problem during the iteration process; Maintain an external archive to store non-dominated solutions; merge the newly generated offspring population with the archive and use Pareto dominance to filter non-dominated solutions; adopt a crowding distance strategy, if the archive has more than 100 solutions, prioritize retaining solutions with larger crowding distances.

[0046] Step 10: Determine if the preset shutdown conditions are met; if not, return to Step 6 and continue the next generation of population optimization iterations; if met, terminate the iteration process. Select 100 solutions from the external archive as the next generation population. After iteration terminates, select the final solution from the archive based on tactical requirements; while satisfying the cooperative time deviation. From the solutions for seconds, select Minimal solution; Cooperative path output: Track point coordinate sequence and velocity planning for 5 UAVs (unified) ); Combat simulation and application process: After the shutdown conditions are met, the obtained Pareto optimal trajectory solution set is output, and the Pareto optimal solution is used as the final collaborative path planning scheme for the flight weapon cluster. This scheme is used for trajectory verification and combat simulation under combat scenarios, thereby realizing the practical application of the flight weapon cluster.

[0047] like Figure 4 , 5 As shown, this invention establishes an equivalent digital terrain model in a three-dimensional combat space, including multiple threatening peaks, no-fly zones, and typical terrain features. The path planning of the flight weapon cluster is carried out in this environment. By comparing the track length, average flight altitude, and threat avoidance effect of different algorithms in this environment, the effectiveness and robustness of the method of this invention under complex dynamic terrain conditions are verified.

[0048] Battlefield space is Five threat peak models are superimposed in a three-dimensional space; the drone parameters are adjusted to a speed of 100m / s and a minimum turning radius of 500m; the five drones fly in coordination according to the planned path, and the threat avoidance status is detected in real time; key indicators such as actual flight path length, average altitude, arrival time deviation, and threat avoidance success rate are recorded, ultimately realizing the practical application of the flight weapon cluster.

[0049] like Figure 3 Through comparative experiments on the ZCAT standard test set and military scenario simulation, this method demonstrates significant advantages. Six sets of comparative figures show the independent running results of the IPPSO algorithm. The final population (discrete points) closely fits the theoretical Pareto front of the ZCAT problem, and the solution uniformly covers the entire front surface without aggregation. It is proved that the incremental SOM manifold guidance mechanism can effectively maintain population diversity and avoid the "premature convergence" problem of traditional MOPSO.

[0050] This invention presents a collaborative path planning method for flight weapon swarms based on guided solution generation swarm optimization. It introduces incremental self-organizing mapping (SOM) into multi-objective particle swarm optimization (MOPSO), explicitly mining the low-dimensional manifold structure of the Pareto solution set (e.g., a 100×1 one-dimensional latent space for bi-objective problems), replacing traditional random sampling. This solves the problems of slow convergence and poor diversity caused by the random selection of pbest / gbest in traditional MOPSO. Furthermore, by combining an online learning mechanism and an adversarial training module, it provides an efficient and robust solution for swarm collaboration in the context of intelligent warfare, with broad military application prospects.

[0051] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for collaborative path planning of flight weapon clusters based on guided solution generation and swarm optimization, characterized in that, Includes the following steps: Step 1: Set up a scenario for collaborative flight path planning of flight weapon clusters; Step 2: Construct a mathematical model for map fusion based on the original flight path terrain model and the equivalent mountain peak terrain model; Step 3: Perform flight path parameterization for the flight weapon cluster and determine the optimization objectives and constraints; Step 4: Utilize self-organizing map manifold approximation to explicitly mine the geometric properties of the Pareto solution set, construct the neuron weight distribution through incremental weight updates, and predict the individual optimal and global optimal guided solutions; combine adaptive particle swarm optimization to iteratively update the path population, and output a path that satisfies the collaborative path planning of the flight weapon cluster.

2. The method for collaborative path planning of flight weapon clusters based on guided solution generation and swarm optimization as described in claim 1, characterized in that, The mathematical model formula for map fusion is as follows: ； in, Point The corresponding mountain height, Point The corresponding terrain elevation.

3. The method for collaborative path planning of flight weapon clusters based on guided solution generation and swarm optimization as described in claim 2, characterized in that, The formula for the height of a mountain peak is: ； in, Represents the x and y coordinates of a point on a horizontal plane; Indicates the number of mountain peaks; Indicates the first The height of the mountain peak; Indicates the first The horizontal and vertical coordinates of the center of the mountain peak on the horizontal plane; , Indicates the first The contour parameters of the mountain peak.

4. The method for collaborative path planning of flight weapon clusters based on guided solution generation and swarm optimization as described in claim 1, characterized in that, Step four specifically includes: Step 41: Initialize the particle swarm; Step 42: Construct an incremental SOM model. In each iteration, the current path population is used as the training input. BMU search, neighborhood function calculation and single-step weight update are performed on each particle in the population in sequence. Set the neighborhood radius and the hidden space dimension; The first The initial weights of the SOM are defined as the weight states of generation t-1; For any path sample in the current generation path population Find its BMU ; Calculate the neighborhood influence of nodes surrounding BMU ; Define a path comprehensive evaluation function; Construct a weight update modulation function based on the path comprehensive evaluation function; Perform a weight update; Step 43: Map the particles in the current population to the model latent space, and predict and generate a guiding solution based on the mapping results to guide the multi-objective particle swarm search. Step 44: Update the particle's velocity and position using the guided solution; Step 45: Update the external archive; Step 46: Stop the iteration when the preset shutdown conditions are met.

5. The method for collaborative path planning of flight weapon clusters based on guided solution generation and swarm optimization as described in claim 4, characterized in that, The formula for the comprehensive evaluation function of the path is: in, For path The average or cumulative flight elevation along the path; , These are the normalization coefficients; These are the weighting coefficients.

6. The method for collaborative path planning of flight weapon clusters based on guided solution generation and swarm optimization as described in claim 5, characterized in that, The formula for the weight update modulation function is: in, This is the modulation intensity parameter.

7. The method for cooperative path planning of flight weapon clusters based on guided solution generation and swarm optimization as described in claim 6, characterized in that, The formula for weight update is: in, It is the first Learning rate.

8. The method for collaborative path planning of flight weapon clusters based on guided solution generation and swarm optimization as described in claim 4, characterized in that, Step 43 specifically includes: Will The weight vector of each neighboring neuron in the decision space The individual optimal guided solution is generated by weighting the average of the reciprocals of the magnitudes of the weight vectors. The formula is: in, These are the weighting coefficients.

9. The method for collaborative path planning of flight weapon clusters based on guided solution generation and swarm optimization as described in claim 8, characterized in that, The formula for the weighting coefficients is: In the formula, The second norm of a vector; It is a preset minimum positive number.

10. A collaborative path planning system for flight weapon clusters based on guided solution generation and swarm optimization, characterized in that: include: Memory is used to store instructions that can be executed by the processor; A processor for executing instructions to implement the collaborative path planning method for flight weapon clusters based on guided solution generation as described in any one of claims 1-9.

Images