Heuristic-driven unmanned aerial vehicle multi-target task planning and route planning system

Abstract

The invention relates to the technical field of unmanned aerial vehicles, in particular to a heuristic driven unmanned aerial vehicle multi-target task planning and flight path planning system. Tasks are decoupled into unmanned aerial vehicle, task and battlefield environment models through the model module, and the model module is in data connection with the optimization algorithm module. And the optimization algorithm module receives the decoupled model data, determines a planning target, generates an optimization result and feeds back the optimization result to the model module. And the path planning module generates an unmanned aerial vehicle flight path according to the optimization result. And the target identification module receives the real-time image data, and detects and identifies a target by using the pre-training model. And the target tracking module receives the identified target information and calculates a prediction trajectory. The algorithm optimization module optimizes algorithm iteration and mutation operators based on historical data and dynamic weight. The system effectively improves task completion efficiency, path optimization, safety and adaptability, and develops new possibility for intelligent application of the unmanned aerial vehicle in a complex dynamic environment.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, in particular to a heuristic-driven unmanned aerial vehicle multi-objective mission planning and trajectory planning system. Background Art

[0002] With the rapid development of UAV technology, multi-target UAV mission planning systems are increasingly used in military reconnaissance, disaster relief, environmental monitoring and other fields. However, facing the increasingly complex mission environment and diversified mission requirements, the existing UAV mission planning systems still have many challenges and shortcomings.

[0003] Traditional UAV mission planning methods usually use a single optimization algorithm, such as genetic algorithm or particle swarm optimization algorithm, to deal with multi-objective planning problems. Although these methods can solve simple mission planning problems to a certain extent, they often perform poorly when dealing with complex multi-objective scenarios. First, a single algorithm is difficult to effectively balance multiple conflicting objectives, such as mission completion rate, flight length, execution time, and path safety. Secondly, these methods have low computational efficiency when facing large-scale missions and complex environments, and it is difficult to meet the needs of real-time planning. In addition, traditional methods lack the ability to adapt to dynamic environments and cannot effectively deal with emergencies and environmental changes that occur during the mission.

[0004] There are also some improved multi-objective optimization methods in the prior art, such as the use of weighted methods or Pareto optimal methods to deal with multi-objective problems. However, these methods still have some inherent limitations. The weighted method requires the weights of each objective to be set in advance, which is often difficult to accurately determine in practical applications. Although the Pareto optimal method can obtain a set of non-dominated solutions, it lacks an effective mechanism for the selection and decision-making of solutions. At the same time, when these methods deal with high-dimensional, large-scale task planning problems, the computational complexity increases sharply, making it difficult to meet the needs of practical applications.

[0005] In terms of path planning, existing technologies mostly use traditional A* algorithms or rapidly expanding random tree (RRT) algorithms. These algorithms perform well in simple environments, but often have difficulty generating optimal paths in complex three-dimensional and dynamic environments. Especially in the presence of multiple threat areas and obstacles, these algorithms have difficulty balancing the requirements of path length and safety.

[0006] In addition, existing UAV mission planning systems usually separate functions such as task allocation, path planning, and target identification, and lack effective integration and coordination mechanisms. This makes it difficult for the system to fully consider the mutual influence of various factors, thus affecting the effectiveness and efficiency of the overall planning. Summary of the invention

[0007] The present invention aims to solve the above technical problems and proposes a heuristic-driven UAV multi-objective mission planning and trajectory planning system. The system achieves efficient planning and optimization of complex multi-objective UAV missions through innovative modular design and collaboration of advanced algorithms.

[0008] The present invention proposes a heuristic-driven UAV multi-objective mission planning and trajectory planning system, comprising:

[0009] Model module, used to:

[0010] Decouple the mission model into UAV model, mission model and battlefield environment model;

[0011] Transmit the decoupled model data to the optimization algorithm module;

[0012] The optimization algorithm module is connected with the model module data and is used to:

[0013] Receiving the decoupled model data sent by the model module;

[0014] Based on the decoupled model data, determining planning objectives and generating optimization results;

[0015] Feeding back the optimization result to the model module;

[0016] The path planning module is data-connected to the optimization algorithm module and is used to:

[0017] Receiving the optimization result generated by the optimization algorithm module;

[0018] Based on the optimization results, generating a flight path for the UAV;

[0019] The target recognition module is connected to the path planning module and is used to:

[0020] receiving real-time image data;

[0021] Detecting and identifying targets from the real-time image data based on a pre-trained target recognition model;

[0022] The target tracking module is data-connected to the target identification module and is used to:

[0023] Receiving target information identified by the target identification module;

[0024] Based on the trajectory prediction model, calculate the target prediction trajectory;

[0025] The algorithm optimization module is connected to the optimization algorithm module data and is used to:

[0026] Based on historical data analysis results and adaptive mutation operators with dynamic weights, the algorithm iteration times and mutation operators are optimized.

[0027] Preferably, the model module comprises:

[0028] A drone model unit, used to store and update the constraints and capabilities data of the drone;

[0029] A task model unit, used to store and update the location, area, timing and priority data of tasks;

[0030] Battlefield environment model unit, used to store and update mission information, threat information, electromagnetic environment information and terrain data in the battlefield area;

[0031] The model decoupling unit is connected to the data of the UAV model unit, the mission model unit and the battlefield environment model unit, and is used to decouple the received data of each unit to generate decoupled model data.

[0032] Preferably, the optimization algorithm module comprises:

[0033] Multi-objective optimization unit, used to construct multi-objective optimization problems based on the MOEA / D (Multi-Objective Evolutionary Algorithmbased on Decomposition) framework;

[0034] A decomposition optimization unit, connected to the multi-objective optimization unit, for decomposing the multi-objective optimization problem into a track optimization sub-problem and a task sequence optimization sub-problem;

[0035] A genetic algorithm unit, connected to the decomposition optimization unit, for solving the track optimization sub-problem and the task sequence optimization sub-problem;

[0036] A non-dominated sorting unit, connected to the genetic algorithm unit data, for performing non-dominated sorting on individuals of the population;

[0037] The elite strategy unit is data-connected to the non-dominated sorting unit and is used to retain and update the elite solution set.

[0038] Preferably, the path planning module comprises:

[0039] The local path planning network is used to perform local path planning for the target point and calculate the flight path of the UAV to reach the target point;

[0040] A global path planning network is used to establish the flight path between the drone and each target point;

[0041] The path optimization unit is connected to the local path planning network and the global path planning network data, and is used to optimize the planning results of the global path planning network based on the output results of the local path planning network.

[0042] Preferably, the target recognition module comprises:

[0043] An image preprocessing unit, used for preprocessing the received real-time image data;

[0044] an attention model unit, data-connected to the image preprocessing unit, and used to extract features from the preprocessed image;

[0045] A target classification unit, connected to the attention model unit data, for classifying the target based on the extracted features;

[0046] The target positioning unit is data-connected with the target classification unit and is used to determine the position of the classified target in the image.

[0047] Preferably, the target tracking module comprises:

[0048] Trajectory prediction network, used to predict the target's future trajectory based on historical target position data;

[0049] A trajectory smoothing unit, connected to the trajectory prediction network data, for smoothing the predicted trajectory;

[0050] The target guidance unit is data-connected to the trajectory smoothing unit and is used to generate a UAV guidance instruction based on the smoothed predicted trajectory.

[0051] Preferably, the algorithm optimization module comprises:

[0052] An iterative optimization unit, used for determining the number of iterative calculations based on the difference between the distance between the UAV and the target calculated each time and the current distance between the UAV and the target;

[0053] The mutation operator optimization unit is used to calculate the difference between the individuals in the population and the individuals in the previous generation and the stability of the individuals in the population itself, and generate an adaptive mutation operator;

[0054] The parameter adaptive unit is data-connected with the iterative optimization unit and the mutation operator optimization unit, and is used to dynamically adjust the parameters in the MOEA / D framework.

[0055] Preferably, the genetic algorithm unit comprises:

[0056] The encoding unit is used to encode the UAV trajectory and task sequence into chromosomes;

[0057] The crossover unit is used to perform SBX (Simulated Binary Crossover) crossover operation on the selected parent chromosome;

[0058] The mutation unit is used to perform PM (Polynomial Mutation) mutation operation on the offspring chromosomes after crossover;

[0059] A fitness calculation unit, used to calculate the multi-objective fitness values ​​of the chromosome;

[0060] The selection unit is used to select excellent individuals to enter the next generation based on fitness values.

[0061] Preferably, it further comprises a constraint processing module, which is data-connected to the optimization algorithm module and is used for:

[0062] Receiving candidate solutions generated by the optimization algorithm module;

[0063] Based on preset constraints, evaluating the feasibility of the candidate solution;

[0064] The evaluation results are fed back to the optimization algorithm module to guide the next round of optimization.

[0065] Preferably, it is characterized by comprising the following steps:

[0066] S1, obtains the constraint and capability model of the UAV and the location, area, timing and priority model of each task;

[0067] S2, taking the shortest UAV track distance and the shortest UAV mission execution time as the optimization objectives and the UAV flight constraints as the constraints, a multi-objective optimization function is established;

[0068] S3, using the MOEA / D framework to build a task optimization model, decomposing the multi-objective optimization function into two sub-problems: track optimization and task sequence optimization;

[0069] S4, using a genetic algorithm to solve the track optimization and task sequence optimization sub-problems, wherein:

[0070] Initialize the population and calculate the multi-objective fitness values ​​of individuals;

[0071] Perform non-dominated sorting to obtain non-dominated sets, external sets and remaining individuals;

[0072] For the remaining individuals, elite selection strategy, SBX crossover and PM mutation are used to generate new individuals;

[0073] Update non-dominated sets and external sets;

[0074] S5. Determine whether the termination condition is met. If not, return to step S4; otherwise, output the optimal solution in the non-dominated set as the planning result of the UAV flight path and task sequence.

[0075] S6. Based on the planning result, use the local path planning network and the global path planning network to generate a detailed UAV flight path.

[0076] S7. During the task execution, use the target recognition module to detect and identify targets in real time, and use the target tracking module to predict the target trajectory and guide the UAV.

[0077] S8. Use the algorithm optimization module to dynamically adjust the number of iterative calculations and mutation operators to improve the algorithm efficiency and adaptability.

[0078] S9. Repeat steps S2 to S8 until all tasks are completed or the preset termination condition is met.

[0079] Advantages of the present invention:

[0080] From a macroscopic perspective, the system of the present invention realizes the comprehensive optimization of multi-objective UAV mission planning. It can not only effectively balance multiple objectives such as task completion rate, flight range length, execution time, and path safety, but also adapt to complex dynamic environments, providing strong technical support for the efficient operation of UAVs in various application scenarios.

[0081] In terms of system architecture, the present invention adopts a modular design, including a model module, an optimization algorithm module, a path planning module, a target recognition module, a target tracking module, and an algorithm optimization module. This design not only improves the flexibility and scalability of the system, but also realizes the effective cooperation between functional modules. For example, the task plan generated by the optimization algorithm module can directly guide the work of the path planning module, and the real-time information of the target recognition and tracking module can be fed back to the optimization algorithm module for dynamic adjustment. This cooperation mechanism significantly improves the overall performance and adaptability of the system.

[0082] At the algorithm level, the present invention cleverly combines a variety of advanced optimization technologies. The MOEA / D framework is used to handle multi-objective optimization problems, effectively solving the deficiencies of traditional methods in target balance and computational efficiency. Improved path planning algorithms, such as the optimized RRT* algorithm, can generate safer and more efficient flight paths in complex three-dimensional environments. Deep learning-based target recognition technology and the target tracking algorithm based on Kalman filtering greatly improve the system's ability to handle dynamic targets.

[0083] It is particularly worth mentioning that the algorithm optimization module of the present invention realizes dynamic adjustment of the optimization process. By analyzing historical data and using adaptive mutation operators, the system can automatically adjust the number of iterations and algorithm parameters according to changes in the task environment and optimization process. This not only improves the convergence speed of the algorithm, but also enhances the system's adaptability to different task types.

[0084] In terms of practical application effects, the system of the present invention shows significant advantages. Compared with traditional methods, it can significantly reduce the total flight distance and average mission execution time while completing more tasks. More importantly, the flight path generated by the system has higher safety, which is crucial for drones performing tasks in complex environments. At the same time, the rapid convergence characteristics of the system enable it to meet the needs of real-time planning, making it possible for the dynamic mission adjustment of drones.

[0085] In summary, the multi-objective UAV mission planning system of the present invention effectively solves many problems existing in the prior art through innovative system design and algorithm integration. It not only achieves comprehensive improvement in task completion efficiency, path optimization, safety and adaptability, but also opens up new possibilities for the intelligent application of UAVs in complex and dynamic environments. The successful application of this system will greatly promote the further development and application of UAV technology in military and civilian fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 is the top-level structure of the system of the present invention;

[0087] Figure 2 It is the internal structure diagram of the model module of the present invention;

[0088] Figure 3 It is the internal structure diagram of the optimization algorithm module of the present invention;

[0089] Figure 4 is an internal structure diagram of the path planning module of the present invention;

[0090] Figure 5 is an internal structure diagram of the target recognition module of the present invention;

[0091] Figure 6 It is the internal structure diagram of the target tracking module of the present invention;

[0092] Figure 7 It is the internal structure diagram of the algorithm optimization module of the present invention; DETAILED DESCRIPTION

[0093] Please refer to the attached Figure 1-7,The present invention provides a heuristic-driven UAV multi-objective mission planning and trajectory planning system. ,The system achieves efficient planning and optimization of complex ,multi-objective UAV missions through innovative modular design and ,cooperation of advanced algorithms.

[0094] The system of the present invention includes a model module, an optimization algorithm module, a path planning module, a target recognition module, a target tracking module and an algorithm optimization module. These modules jointly complete the complex process of multi-target UAV mission planning through close data connection and interaction.

[0095] Model module 1 is responsible for decoupling the mission model into UAV model, mission model and battlefield environment model. This decoupling design enhances the flexibility and scalability of the system. For example, the UAV model can include parameters such as flight speed, endurance, and load capacity; the mission model can include information such as target location, priority, and time window; and the battlefield environment model can include data such as terrain, weather conditions, and threat distribution. Model module 1 transmits these decoupled model data to optimization algorithm module 2 to provide a basis for subsequent optimization calculations.

[0096] The optimization algorithm module 2 is the core of this system. It receives the decoupled model data sent by the model module 1, determines the planning objectives based on these data, and generates optimization results. The optimization algorithm module 2 adopts a multi-objective optimization method, taking the shortest UAV track distance and the shortest mission execution time as the optimization objectives. This multi-objective design is more in line with actual application needs and can strike a balance between task efficiency and resource utilization.

[0097] Preferably, the optimization algorithm module 2 adopts the MOEA / D (Multi-Objective Evolutionary Algorithmbased on Decomposition) framework to construct and solve the multi-objective optimization problem. The core idea of ​​the MOEA / D algorithm is to decompose the multi-objective optimization problem into a series of scalar optimization sub-problems, and then optimize these sub-problems simultaneously. In the present invention, the MOEA / D algorithm can be expressed as:

[0098] minF(x)=(f 1 (x),f 2 (x),...,f m (x)),

[0099] Among them, F(x) is the multi-objective optimization function, f i (x) is the ith objective function, and m is the number of objective functions. In the present invention, m=2, corresponding to the track distance and the mission time, respectively.

[0100] The path planning module 3 receives the optimization results generated by the optimization algorithm module 2, and generates a specific flight path for the drone based on these results. In one embodiment of the present invention, the path planning module 3 can use an improved A* algorithm to generate the path. The improved A* algorithm can be expressed as:

[0101] f(n)=g(n)+h(n)+r(n),

[0102] Among them, f(n) is the evaluation function of node n, g(n) is the actual cost from the starting node to n, h(n) is the estimated cost from n to the target node, and r(n) is an additional term considering task priority and time window.

[0103] The target recognition module 4 and the target tracking module 5 work together to achieve real-time recognition and tracking of the target. The target recognition module 4 uses a pre-trained deep learning model, such as YOLOv5 or Faster R-CNN, to detect and recognize the target from real-time image data. The target tracking module 5 uses algorithms such as Kalman filter or particle filter to calculate the predicted trajectory of the target based on the target recognition result.

[0104] The algorithm optimization module 6 is another innovative point of the present invention. It optimizes the algorithm's iterative calculation times and mutation operators by analyzing historical data and using adaptive mutation operators with dynamic weights. This adaptive optimization mechanism can significantly improve the efficiency and robustness of the algorithm. For example, the adaptive mutation operator can be expressed as:

[0105]

[0106] Among them, F is the adaptive mutation operator, α is the weight coefficient, Δ is the population difference, σ g is population stability, F 0 is the initial mutation operator.

[0107] The model module 1 further includes a drone model unit 11, a mission model unit 12, a battlefield environment model unit 13 and a model decoupling unit 14. This subdivision design enables the system to process various types of data more finely, thereby improving the accuracy and adaptability of the model.

[0108] The UAV model unit 11 stores and updates the constraint and capability data of the UAV. For example, it may include parameters such as the maximum flight speed (e.g., 50m / s), maximum flight altitude (e.g., 5000m), and maximum load (e.g., 10kg) of the UAV. These parameters play a key role in the mission planning process, ensuring that the generated mission plan is within the capabilities of the UAV.

[0109] The task model unit 12 stores and updates the location, area, timing and priority data of the task. For example, the task location can be represented by latitude and longitude coordinates, the task area can be described by a polygonal area, the timing can include the start and end time windows of the task, and the priority can be represented by an integer from 1 to 10. These detailed task information enables the system to generate more accurate and efficient task planning solutions.

[0110] The battlefield environment model unit 13 stores and updates the mission information, threat information, electromagnetic environment information and terrain data in the battlefield area. For example, threat information may include the location and coverage of the enemy's air defense system, electromagnetic environment information may include the communication interference area, and terrain data may include the distribution of natural obstacles such as mountains and rivers. Consideration of these environmental information makes the mission plan generated by the system safer and more reliable.

[0111] The model decoupling unit 14 is the core of the model module 1, which is responsible for decoupling the received unit data and generating decoupled model data. This decoupling process enables the system to process various types of data more flexibly, thereby improving the scalability and adaptability of the system.

[0112] The optimization algorithm module 2 includes a multi-objective optimization unit 21, a decomposition optimization unit 22, a genetic algorithm unit 23, a non-dominated sorting unit 24 and an elite strategy unit 25. This modular design enables the system to efficiently handle complex multi-objective optimization problems.

[0113] The multi-objective optimization unit 21 constructs a multi-objective optimization problem based on the MOEA / D framework. In one embodiment of the present invention, the multi-objective optimization problem can be expressed as:

[0114] minF(x)=(f 1 (x),f 2 (x)),

[0115] Among them, f 1 (x) represents the track distance, f 2 (x) represents the task execution time.

[0116] The decomposition optimization unit 22 decomposes the multi-objective optimization problem into a track optimization sub-problem and a task sequence optimization sub-problem. This decomposition strategy can significantly reduce the complexity of the problem and improve the efficiency of solving. For example, the track optimization sub-problem can be expressed as:

[0117]

[0118] Among them, d(x i ,x i+1 ) represents the distance between adjacent task points, and n is the number of task points.

[0119] The genetic algorithm unit 23 is responsible for solving the track optimization subproblem and the task sequence optimization subproblem. In the present invention, the chromosome encoding of the genetic algorithm can adopt the real number encoding mode, the crossover operation can adopt the simulated binary crossover (SBX), and the mutation operation can adopt the polynomial mutation (PM). These advanced genetic operators can improve the search efficiency of the algorithm and the quality of the solution.

[0120] The non-dominated sorting unit 24 performs non-dominated sorting on the individuals in the population. In multi-objective optimization problems, non-dominated sorting is an important individual evaluation method. The present invention adopts a fast non-dominated sorting algorithm, and its time complexity is O(MN^2), where M is the number of targets and N is the population size.

[0121] The elite strategy unit 25 is responsible for retaining and updating the elite solution set. The elite strategy can ensure that the excellent individuals in the population will not be lost during the evolution process. In the present invention, the size of the elite solution set can be set to 10% of the population size. This ratio can achieve a good balance between maintaining population diversity and improving the algorithm convergence speed.

[0122] Through the collaborative work of the above modules, the system of the present invention can efficiently handle complex multi-target UAV mission planning problems and generate optimized mission execution plans. The system not only takes into account the performance constraints and mission requirements of the UAV, but also fully considers the impact of the battlefield environment, and has strong practicality and adaptability. In addition, the modular design of the system and the application of advanced algorithms give it good scalability and optimization potential, laying the foundation for future functional expansion and performance improvement.

[0123] The system of the present invention further comprises a path planning module 3, which comprises a local path planning network 31, a global path planning network 32 and a path optimization unit 33. This multi-level path planning structure can effectively handle complex mission environments and generate efficient and feasible flight paths.

[0124] The local path planning network 31 is responsible for local path planning of the target point and calculating the flight path of the drone to reach the target point. In one embodiment of the present invention, the local path planning network 31 can adopt an improved rapidly expanding random tree (RRT*) algorithm. The core idea of ​​the algorithm is to randomly sample in the state space and build a tree by connecting the sampled points, and finally find a feasible path from the starting point to the target point. Preferably, the improved version of the RRT* algorithm can be expressed as:

[0125]

[0126] Among them, cost(x new ) represents the cost of the new node, x near represents the set of nodes near the new node, c(xnear ,x new ) represents the cost from the neighboring node to the new node. This improvement can significantly improve the quality of the path and the convergence speed of the algorithm.

[0127] The global path planning network 32 is used to establish the flight path between the drone and each target point. In a preferred embodiment of the present invention, the global path planning network 32 adopts an improved A* algorithm. Compared with the traditional A* algorithm, the improved version introduces a dynamic weight factor, which can dynamically adjust the weight of the heuristic function according to the urgency of the task and the complexity of the environment. The improved A* algorithm can be expressed as:

[0128] f(n)=g(n)+w(n)·h(n),

[0129] Where w(n) is a dynamic weight factor, and its value range is usually [1,2]. When w(n) = 1, the algorithm degenerates into the traditional A" algorithm; when w(n)>1, the algorithm will be more inclined to explore nodes closer to the target, thereby speeding up the search.

[0130] The path optimization unit 33 is connected to the local path planning network 31 and the global path planning network 32, and is used to optimize the planning results of the global path planning network based on the output results of the local path planning network. Preferably, the path optimization unit 33 can use a B-spline curve smoothing algorithm to optimize the generated path to improve the smoothness and flyability of the path. The B-spline curve can be expressed as:

[0131]

[0132] Among them, N i,p (t) is the p-order B-spline basis function, P i By adjusting the position of the control points, a smoother and more efficient flight path can be obtained.

[0133] The target recognition module 4 of the present invention includes an image preprocessing unit 41, an attention model unit 42, a target classification unit 43 and a target positioning unit 44. This modular design enables the system to efficiently and accurately recognize and locate targets.

[0134] The image preprocessing unit 41 preprocesses the received real-time image data. In one embodiment of the present invention, the preprocessing process may include operations such as image scaling, histogram equalization, and data normalization. For example, the image may be scaled to 224x224 pixels to meet the input requirements of the subsequent deep learning model. Histogram equalization can improve the contrast of the image and help identify the target. Data normalization can normalize the pixel value to the range of [0, 1], which is conducive to the training and reasoning of the neural network.

[0135] The attention model unit 42 is data-connected to the image preprocessing unit 41, and is used to extract features from the preprocessed image. The present invention preferably uses a self-attention mechanism to extract image features. The self-attention mechanism can be expressed as:

[0136]

[0137] Where Q, K, and V represent query, key, and value matrices, respectively. k is the dimension of the key vector. This mechanism can capture long-range dependencies in the image and improve the effect of feature extraction.

[0138] The target classification unit 43 is data-connected to the attention model unit 42 for classifying targets based on the extracted features. In a preferred embodiment of the present invention, the target classification unit 43 adopts a convolutional neural network (CNN) structure, such as ResNet-50 or EfficientNet-B0. These advanced CNN models can achieve high-precision target classification while maintaining low computational complexity. For example, the core idea of ​​ResNet-50 is to introduce residual connections, which can be expressed as:

[0139] H(x)=F(x)+x,

[0140] Among them, H(x) is the desired underlying mapping, F(x) is the residual function, and x is the input. This structure can effectively alleviate the gradient vanishing problem of deep networks and improve the performance of the model.

[0141] The target positioning unit 44 is data-connected to the target classification unit 43 to determine the position of the classified target in the image. The system of the present invention adopts a region-based target detection algorithm, such as Faster R-CNN or YOLO. Taking Faster R-CNN as an example, its target positioning process can be divided into two stages: the region proposal network (RPN) generates candidate regions, and then these candidate regions are accurately classified and bounding box regressed. Bounding box regression can be expressed as:

[0142] t x =(xx a ) / w a ,

[0143] t y =(yy a ) / h a ,

[0144] t w =log(w / w a ),

[0145] t h=log(h / h a ),

[0146] Among them, (x, y, w, h) represents the predicted bounding box parameters, (x a ,y a , w a ,h a ) represents the anchor box parameters.

[0147] The target tracking module 5 of the present invention includes a trajectory prediction network 51, a trajectory smoothing unit 52 and a target guidance unit 53. This structural design enables the system to accurately predict the target motion and provide effective guidance information for the drone.

[0148] The trajectory prediction network 51 predicts the future trajectory of the target based on the historical target position data. In one embodiment of the present invention, the trajectory prediction network 51 adopts a long short-term memory network (LSTM) structure. The core of LSTM is its memory unit, which can be expressed as:

[0149] f t =σ(W f ·[h t-1 , x t ]+b f ),

[0150] i t =σ(W i ·[h t-1 , x t ]+b i ),

[0151]

[0152] o t =σ(W o ·[h t-1 , x t ]+b o ),

[0153] h t =o t *tanh(C t ),

[0154] Among them, f t 、i t , o t Respectively represent the forget gate, input gate and output gate, C t Indicates the unit state, h t Represents the hidden state. This structure can effectively capture the temporal characteristics of target motion and improve the accuracy of trajectory prediction.

[0155] The trajectory smoothing unit 52 is data-connected to the trajectory prediction network 51 for smoothing the predicted trajectory. The present invention preferably uses a Kalman filter for trajectory smoothing. The core equation of the Kalman filter includes a prediction equation and an update equation:

[0156] Prediction equation:

[0157]

[0158] Update equation:

[0159]

[0160]

[0161] in, represents the state estimation, P k represents the estimation error covariance, K k Represents the Kalman gain. Through this filtering process, the noise and fluctuation in trajectory prediction can be effectively reduced.

[0162] The target guidance unit 53 is data-connected to the trajectory smoothing unit 52, and is used to generate the guidance instruction of the UAV based on the smoothed predicted trajectory. In a preferred embodiment of the present invention, the target guidance unit 53 uses the proportional navigation method (PN) to generate the guidance instruction. The acceleration instruction of the PN method can be expressed as:

[0163] a c =NV c λ,

[0164] Where N is the guidance coefficient (usually 3-5), V c is the approach speed, and λ is the line-of-sight angular velocity. This method can effectively intercept moving targets while being computationally simple.

[0165] The algorithm optimization module 6 of the present invention comprises an iterative optimization unit 61, a mutation operator optimization unit 62 and a parameter adaptive unit 63. This structural design enables the system to dynamically adjust the algorithm parameters, thereby improving the optimization efficiency and adaptability.

[0166] The iterative optimization unit 61 determines the number of iterative calculations based on the difference between the distance between the drone and the target each time and the current distance between the drone and the target. In one embodiment of the present invention, the number of iterations can be determined using an adaptive step length method, which is expressed as:

[0167]

[0168] Among them, T k represents the number of iterations of the kth iteration, d krepresents the target distance of the kth iteration, and α is the adjustment coefficient (usually 0.1-0.5). This method can dynamically adjust the number of iterations according to the convergence of the optimization process and improve the efficiency of the algorithm.

[0169] The mutation operator optimization unit 62 is used to calculate the difference between the population individuals and the previous generation population individuals and the stability of the population individuals themselves, and generate an adaptive mutation operator. Preferably, the adaptive mutation operator can be expressed as:

[0170]

[0171] Among them, F m is the adaptive mutation operator, F base is the basic mutation rate (usually 0.01-0.05), β is the weight coefficient (usually 0.1-0.5), Δ is the population difference, and σ is the population stability. This adaptive mechanism can strike a balance between maintaining population diversity and improving convergence speed.

[0172] The parameter adaptive unit 63 is data-connected with the iterative optimization unit 61 and the mutation operator optimization unit 62 to dynamically adjust the parameters in the MOEA / D framework. For example, the weight vector in the MOEA / D can be adaptively adjusted according to the current Pareto frontier distribution:

[0173]

[0174] Among them, w i represents the i-th weight vector, represents the ideal weight distribution, and γ is the learning rate (usually 0.01-0.1). This adaptive adjustment can make the algorithm better adapt to the characteristics of the problem and improve the quality of the solution.

[0175] Through the collaborative work of the above modules, the system of the present invention can achieve efficient and accurate multi-target UAV mission planning. The system not only takes into account the optimization of path planning, but also integrates advanced target recognition and tracking technology, and realizes adaptive optimization through the algorithm optimization module. This comprehensive design enables the system to perform well in complex and dynamic mission environments, providing strong support for the intelligent application of UAVs.

[0176] The genetic algorithm unit 23 of the present invention further includes an encoding unit 231, a crossover unit 232, a mutation unit 233, a fitness calculation unit 234 and a selection unit 235. This sophisticated module design enables the genetic algorithm to handle complex optimization problems more efficiently.

[0177] The encoding unit 231 is responsible for encoding the UAV track and task sequence into chromosomes. In a preferred embodiment of the present invention, a real number encoding method is adopted. For example, for a track containing n task points, it can be represented by a 2n-dimensional real number vector:

[0178] C=(x 1 ,y 1 ,x 2 ,y 2 ,...,x n ,y n ),

[0179] Among them, (x i ,y i ) represents the coordinates of the i-th task point. This encoding method is intuitive and simple, and is convenient for subsequent genetic operations.

[0180] The crossover unit 232 is used to perform a SBX (Simulated Binary Crossover) crossover operation on the selected parent chromosome. The SBX crossover operation can be expressed as:

[0181] C 1,k =0.5[(1+β k ) 1，k +(1-β k ) 2,k ],

[0182] C 2,k =0.5[(1-β k ) 1，k +(1+β k ) 2,k ],

[0183] Among them, c 1,k and c 2,k is the kth gene of the offspring chromosome, p 1,k and p 2,k is the κth gene of the parent chromosome, β k is a random number whose distribution is controlled by a parameter η. Usually, η takes a value between 20 and 100. This crossover operation can well maintain the characteristics of the parent generation while generating sufficient diversity.

[0184] The mutation unit 233 is used to perform a PM (Polynomial Mutation) mutation operation on the offspring chromosome after the crossover. The PM mutation operation can be expressed as:

[0185] c k =p k +δ k (u k -l k ),

[0186] Among them, c k is the gene value after mutation, p k is the gene value before mutation, u k and l k are the upper and lower bounds of the gene, δ k is a small perturbation whose distribution is controlled by a parameter η_m. Usually, η_m takes a value between 20 and 100. This mutation operation can maintain local search capabilities while also having a certain probability of producing larger changes.

[0187] The fitness calculation unit 234 is used to calculate the multi-objective fitness value of the chromosome. In the present invention, the fitness function includes two objectives: track length and task completion time. It can be expressed as:

[0188]

[0189] Among them, d(x i , x i+1 ) represents the distance between adjacent task points, t i represents the residence time at the i-th mission point, and v represents the average flight speed of the UAV.

[0190] The selection unit 235 is used to select excellent individuals to enter the next generation based on the fitness value. The present invention adopts an elite selection strategy based on non-dominated sorting. First, the population is non-dominated sorted, and then individuals are selected from high to low according to the non-dominated level until the population size reaches a preset value. When all individuals of a certain non-dominated level cannot be selected, the crowding ranking is used to select individuals in this level. This selection strategy ensures the retention of elite individuals and maintains the diversity of the population.

[0191] The system of the present invention further comprises a constraint processing module 7, which is data-connected to the optimization algorithm module 2. The introduction of the constraint processing module 7 enables the system to better handle various constraint conditions in practical applications, thereby improving the feasibility of the optimization result.

[0192] The constraint processing module 7 first receives the candidate solution generated by the optimization algorithm module 2. In one embodiment of the present invention, the candidate solution can be represented as a structure including a task sequence and a corresponding track:

[0193] S={(T 1 ,P 1 ),(T 2 ,P 2 ),...,(T n ,P n )},

[0194] Among them, T i represents the i-th task, Pi Indicates the track that completes the task.

[0195] Then, the constraint processing module 7 evaluates the feasibility of the candidate solution based on the preset constraint conditions. The constraint conditions considered in the present invention mainly include:

[0196] 1. UAV endurance constraints:

[0197]

[0198] Among them, T max The maximum flight time of the drone.

[0199] 2. Task time window constraints:

[0200]

[0201] Among them, $t_{start,i}$ and $t_{end,i}$ are the start and end time windows of the i-th task, respectively, and $t_{arrive,i}$ is the time when the UAV arrives at the i-th task point.

[0202] 3. Track safety constraints:

[0203]

[0204] Among them, d(P,O) represents the distance from the track point P to the obstacle, d safe is the safety distance threshold.

[0205] The constraint processing module 7 feeds back the evaluation results to the optimization algorithm module 2 for guiding the next round of optimization. Preferably, the penalty function method is used to process the constraints, and the degree of constraint violation is converted into a penalty term and added to the objective function:

[0206]

[0207] Among them, F(x) is the original objective function, g i (x) is the i-th constraint function, r i is the corresponding penalty coefficient. The influence of the constraints can be gradually increased during the optimization process, and finally a solution that satisfies all constraints is obtained.

[0208] The present invention also provides a multi-objective UAV mission planning method based on a heuristic optimization algorithm. The method includes multiple steps, each of which is carefully designed to achieve efficient and reliable mission planning.

[0209] Step S1 obtains the constraint and capability model of the UAV and the location, area, timing and priority model of each task. This step provides the necessary basic data for the subsequent optimization process. For example, the constraint and capability model of the UAV may include parameters such as maximum flight speed, maximum flight altitude, and maximum load, while the task model contains information such as task coordinates, execution time window, and task priority.

[0210] In step S2, the present invention takes the shortest UAV track distance and the shortest UAV mission execution time as optimization objectives, and takes the flight constraints of the UAV as constraints to establish a multi-objective optimization function. This multi-objective design can better balance task efficiency and resource utilization. The optimization function can be expressed as:

[0211] minF(x)=(f 1 (x),f 2 (x)),

[0212] stg i (x)≤0,i=1,2,...,m,

[0213] Among them, f 1 (x) represents the track distance, f 2 (x) represents the task execution time, g i (x) represents various constraints.

[0214] Step S3 uses the MOEA / D framework to build a task optimization model and decompose the multi-objective optimization function into two sub-problems: track optimization and task sequence optimization. The core idea of ​​MOEA / D is to transform the multi-objective problem into a series of single-objective sub-problems, each of which corresponds to a weight vector. For example, the Chebyshev decomposition method can be used:

[0215]

[0216] Among them, w is the weight vector, z * For the ideal point.

[0217] In step S4, the present invention uses a genetic algorithm to solve the sub-problems of trajectory optimization and task sequence optimization. This process includes initializing the population, calculating the multi-objective fitness value of the individual, performing non-dominated sorting, applying the elite selection strategy, SBX crossover and PM mutation, etc. In particular, the non-dominated sorting adopts a fast non-dominated sorting algorithm, and its time complexity is O(MN^2), where M is the number of targets and N is the population size.

[0218] Step S5 determines whether the termination condition is met. Common termination conditions include reaching the maximum number of iterations, reaching the preset calculation time, and no significant improvement in the optimal solution for multiple consecutive generations. If the termination condition is not met, return to step S4 to continue iterating; otherwise, output the optimal solution in the non-dominated set as the planning result of the UAV track and task sequence.

[0219] In step S6, the present invention generates a detailed UAV flight path based on the planning results using a local path planning network and a global path planning network. The local path planning uses an improved RRT* algorithm, while the global path planning uses an improved A* algorithm. This multi-level path planning strategy can effectively handle complex mission environments.

[0220] Step S7: During the task execution, the target recognition module is used to detect and recognize the target in real time, and the target tracking module is used to predict the target trajectory and guide the UAV. The target recognition module uses a deep learning model such as YOLOv5 or FasterR-CNN, while the target tracking module uses a Kalman filter or a particle filter.

[0221] In step S8, the present invention uses an algorithm optimization module to dynamically adjust the number of iterative calculations and mutation operators to improve algorithm efficiency and adaptability. For example, the number of iterations can be dynamically adjusted according to the convergence of the optimization process:

[0222]

[0223] Among them, T k represents the number of iterations of the kth iteration, f(x k ) represents the objective function value of the kth iteration, and α is the adjustment coefficient.

[0224] Finally, step S9 repeats steps S2 to S8 until all tasks are completed or the preset termination condition is met. This cyclic optimization strategy enables the system to adapt to the dynamically changing task environment and continuously improve the planning scheme.

[0225] Through the above steps, the method of the present invention can efficiently handle complex multi-target UAV mission planning problems and generate optimized mission execution plans. The method comprehensively considers UAV performance, mission characteristics and environmental constraints, and has strong practicality and adaptability. At the same time, the method contains a number of innovative technologies, such as MOEA / D framework, improved path planning algorithm, deep learning based target recognition and tracking, adaptive optimization strategy, etc., which make it have significant advantages in efficiency and effect.

[0226] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Heuristic-driven UAV multi-objective mission planning and trajectory planning system, characterized by: include: Model module, used to: Decouple the mission model into UAV model, mission model and battlefield environment model; Transmit the decoupled model data to the optimization algorithm module; The optimization algorithm module is connected with the model module data and is used to: Receiving the decoupled model data sent by the model module; Based on the decoupled model data, determining planning objectives and generating optimization results; Feeding back the optimization result to the model module; The path planning module is data-connected to the optimization algorithm module and is used to: Receiving the optimization result generated by the optimization algorithm module; Based on the optimization results, generating a flight path for the UAV; The target recognition module is connected to the path planning module and is used to: receiving real-time image data; Detecting and identifying targets from the real-time image data based on a pre-trained target recognition model; The target tracking module is data-connected to the target identification module and is used to: Receiving target information identified by the target identification module; Based on the trajectory prediction model, calculate the target prediction trajectory; The algorithm optimization module is connected to the optimization algorithm module data and is used to: Based on historical data analysis results and adaptive mutation operators with dynamic weights, the algorithm iteration times and mutation operators are optimized.

2. The system according to claim 1, characterized in that The model module includes: A drone model unit, used to store and update the constraints and capabilities data of the drone; A task model unit, used to store and update the location, area, timing and priority data of tasks; Battlefield environment model unit, used to store and update mission information, threat information, electromagnetic environment information and terrain data in the battlefield area; The model decoupling unit is connected to the data of the UAV model unit, the mission model unit and the battlefield environment model unit, and is used to decouple the received data of each unit to generate decoupled model data.

3. The system according to claim 1, characterized in that The optimization algorithm module includes: Multi-objective optimization unit, used to construct multi-objective optimization problems based on the MOEA / D framework; A decomposition optimization unit, connected to the multi-objective optimization unit, for decomposing the multi-objective optimization problem into a track optimization sub-problem and a task sequence optimization sub-problem; A genetic algorithm unit, connected to the decomposition optimization unit, for solving the track optimization sub-problem and the task sequence optimization sub-problem; A non-dominated sorting unit, connected to the genetic algorithm unit data, for performing non-dominated sorting on individuals of the population; The elite strategy unit is data-connected to the non-dominated sorting unit and is used to retain and update the elite solution set.

4. The system according to claim 1, characterized in that The path planning module includes: The local path planning network is used to perform local path planning for the target point and calculate the flight path of the UAV to reach the target point; A global path planning network is used to establish the flight path between the drone and each target point; The path optimization unit is connected to the local path planning network and the global path planning network data, and is used to optimize the planning results of the global path planning network based on the output results of the local path planning network.

5. The system according to claim 1, characterized in that The target recognition module comprises: An image preprocessing unit, used for preprocessing the received real-time image data; an attention model unit, data-connected to the image preprocessing unit, and used to extract features from the preprocessed image; A target classification unit, connected to the attention model unit data, for classifying the target based on the extracted features; The target positioning unit is data-connected with the target classification unit and is used to determine the position of the classified target in the image.

6. The system according to claim 1, characterized in that The target tracking module comprises: Trajectory prediction network, used to predict the target's future trajectory based on historical target position data; A trajectory smoothing unit, connected to the trajectory prediction network data, for smoothing the predicted trajectory; The target guidance unit is data-connected to the trajectory smoothing unit and is used to generate a UAV guidance instruction based on the smoothed predicted trajectory.

7. The system according to claim 1, characterized in that The algorithm optimization module includes: An iterative optimization unit, used for determining the number of iterative calculations based on the difference between the distance between the UAV and the target calculated each time and the current distance between the UAV and the target; The mutation operator optimization unit is used to calculate the difference between the individuals in the population and the individuals in the previous generation and the stability of the individuals in the population itself, and generate an adaptive mutation operator; The parameter adaptive unit is data-connected with the iterative optimization unit and the mutation operator optimization unit, and is used to dynamically adjust the parameters in the MOEA / D framework.

8. The system according to claim 3, characterized in that The genetic algorithm unit comprises: The encoding unit is used to encode the UAV trajectory and task sequence into chromosomes; The crossover unit is used to perform SBX crossover operation on the selected parent chromosome; The mutation unit is used to perform PM mutation operations on the offspring chromosomes after crossover; A fitness calculation unit, used to calculate the multi-objective fitness values ​​of the chromosome; The selection unit is used to select excellent individuals to enter the next generation based on fitness values.

9. The system according to claim 1, characterized in that It also includes a constraint processing module, which is data-connected to the optimization algorithm module and is used to: Receiving candidate solutions generated by the optimization algorithm module; Based on preset constraints, evaluating the feasibility of the candidate solution; The evaluation results are fed back to the optimization algorithm module to guide the next round of optimization.

10. A multi-objective UAV mission planning method based on a heuristic optimization algorithm, characterized in that: The following steps are involved: S1, obtains the constraint and capability model of the UAV and the location, area, timing and priority model of each task; S2, taking the shortest UAV track distance and the shortest UAV mission execution time as the optimization objectives and the UAV flight constraints as the constraints, a multi-objective optimization function is established; S3, using the MOEA / D framework to build a task optimization model, decomposing the multi-objective optimization function into two sub-problems: track optimization and task sequence optimization; S4, using a genetic algorithm to solve the track optimization and task sequence optimization sub-problems, wherein: Initialize the population and calculate the multi-objective fitness values ​​of individuals; Perform non-dominated sorting to obtain non-dominated sets, external sets and remaining individuals; For the remaining individuals, elite selection strategy, SBX crossover and PM mutation are used to generate new individuals; Update non-dominated sets and external sets; S5, judging whether the termination condition is met, if not, returning to step S4, otherwise outputting the optimal solution in the non-dominated set as the planning result of the UAV trajectory and task sequence; S6, based on the planning results, using a local path planning network and a global path planning network to generate a detailed UAV flight path; S7, during the mission execution, uses the target recognition module to detect and identify the target in real time, and uses the target tracking module to predict the target trajectory and guide the UAV; S8, using the algorithm optimization module to dynamically adjust the number of iterative calculations and mutation operators to improve algorithm efficiency and adaptability; S9, repeating steps S2 to S8 until all tasks are completed or a preset termination condition is met.

Images