Unmanned aerial vehicle real-time target positioning method based on three-dimensional electromagnetic map

Abstract

The application belongs to the technical field of unmanned aerial vehicle positioning and electromagnetic map fusion, and discloses a real-time target positioning method for unmanned aerial vehicles based on a three-dimensional electromagnetic map, which comprises the following steps: constructing a three-dimensional electromagnetic map based on a Longley-Rice propagation model and GMTED2010 elevation data, and deploying multiple unmanned aerial vehicles to collect target RSS observation values; using an improved wolf swarm algorithm based on fusion reverse learning to replace a traditional MMSE algorithm, and through elite reverse learning to initialize a population, designing a double-item fusion fitness function, dynamically regulating a population update mode, combining unmanned aerial vehicle cooperative control and a kmeans clustering local optimal avoidance strategy, and completing three-dimensional target real-time positioning. The system comprises a map construction module, an unmanned aerial vehicle deployment module, a positioning processing module, a cooperative control module and a result output module.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of UAV positioning and electromagnetic detection technology, and particularly relates to a real-time target positioning method and system for UAVs based on a three-dimensional electromagnetic map. Background Technology

[0002] In the field of UAV positioning research, using UAVs as dynamic receiving nodes to conduct target search and positioning based on received signal strength (RSS) is one of the important technical approaches to solving the positioning challenges in complex environments. It has high engineering application value in various fields such as emergency rescue, military reconnaissance, and wildlife tracking. Compared to traditional static base station positioning, UAVs have a significant advantage in high mobility. After each position estimation, they can dynamically move towards the target coordinates, thereby improving the signal-to-noise ratio of subsequent signal observations. They can even reach areas near the target to directly verify the positioning results, further enhancing the flexibility and practical application value of positioning technology.

[0003] RSS measurement technology boasts advantages such as low hardware cost, low power consumption, and simple requirements for terminal equipment, making it widely used in UAV positioning and the most prevalent signal measurement method in this field. However, this technology has significant limitations. The propagation of radio waves is easily affected by environmental factors such as terrain layout and vegetation cover, and positioning accuracy is highly dependent on the stability of the propagation environment. In areas with complex terrain conditions, such as mountainous regions, using an idealized free-space propagation model for signal prediction can lead to significant deviations between RSS measurements and actual signal strength, resulting in decreased positioning accuracy and even potential target search failure. This limits the engineering application of UAV RSS positioning technology in complex terrain scenarios.

[0004] To address the issues of inaccurate radio wave propagation prediction and insufficient positioning accuracy in complex terrain, radio maps (also known as electromagnetic maps), which accurately describe the characteristics of radio wave propagation, have gradually become an important technological support in this field. Currently, there are three main methods for constructing radio maps in the industry: Measurement acquisition methods require drawing maps from a large number of randomly distributed location measurement samples. While offering high accuracy, these methods suffer from problems such as large deployment of sensors, high construction costs, and low efficiency. Model-free methods rely on a small number of measurement samples and fit maps using machine learning algorithms. These methods offer high construction efficiency, but their performance is easily affected by the rationality of sensor placement and measurement errors, resulting in poor stability. Model-based methods rely on radio wave propagation models and terrain data to estimate radio maps. They are less sensitive to sensor placement and offer high stability, making them the mainstream technology solution for commercial radio map tools. In complex terrain positioning scenarios, model-based methods typically use the Longley-Rice propagation model, combined with real terrain elevation data, to generate high-precision radio maps adapted to irregular and complex terrain, providing reliable prior information to improve positioning accuracy.

[0005] Current research indicates that radio mapping technology has been gradually applied to UAV communication, playing a crucial role in UAV trajectory planning, anti-jamming communication path optimization, and wireless power supply system design, effectively enhancing the adaptability of UAV systems in complex environments. Simultaneously, research on UAV target search and localization technology has also made some progress. Researchers have combined Bayes' theorem, control theory, machine learning, and other techniques to propose various localization schemes, which have been applied to practical scenarios such as reconnaissance, emergency rescue, and wildlife tracking. However, research has found that most existing studies on UAV target localization and search still do not incorporate radio mapping knowledge into the localization process to provide assistance, indicating significant research value in their integrated application. Although both radio mapping and UAV localization have made progress, their deep integration has not been fully explored and applied, failing to fully utilize the terrain influence information contained in radio maps to further improve the accuracy and efficiency of UAV target search and localization in complex terrain.

[0006] Some existing studies employ the Minimum Mean Square Error Estimator (MMSE) for UAV target localization. This algorithm estimates position through statistical inference on a discrete grid space, showing some applicability in two-dimensional localization scenarios, and its logic is simple and easy to implement. However, this approach still has significant technical limitations: firstly, it can only achieve localization in two-dimensional scenarios, failing to consider the impact of the height dimension on the localization results, making it difficult to meet the localization requirements of complex three-dimensional terrain; secondly, when the localization dimension expands from two-dimensional to three-dimensional, the number of discrete grids increases cubically. Due to the limitation of the MMSE algorithm's full grid traversal matching logic, its computational complexity increases significantly, and its operational efficiency decreases markedly, making it difficult to meet the real-time localization requirements of UAVs.

[0007] Based on existing research, current UAV RSS positioning technology still has several technical shortcomings and research problems: First, in complex terrain, the application of idealized propagation models leads to insufficient RSS positioning accuracy, while the integrated application of radio maps and UAV positioning has not been fully explored; second, existing MMSE estimation algorithms are difficult to adapt to 3D positioning scenarios, suffering from high computational complexity and poor real-time performance, making it difficult to meet the actual needs of high-precision and high-real-time positioning in complex terrain. To address these shortcomings, this invention proposes a real-time UAV target positioning method and system based on a 3D electromagnetic map. It integrates a back-learning method to replace the traditional MMSE estimation algorithm based on the improved wolf pack algorithm (AGWO) proposed by Ma et al. (2022). Relying on a 3D electromagnetic map constructed based on the Longley-Rice propagation model and real terrain data, it achieves high-precision, low-complexity real-time positioning of UAVs in complex terrain, solving the core problems of existing technologies and promoting the engineering application and popularization of related technologies. Summary of the Invention

[0008] Addressing the limitations of existing positioning technologies, such as the limited adaptability of 2D positioning schemes and the increased computational complexity and insufficient real-time performance of the minimum mean square error (MMSE) estimation algorithm in 3D positioning scenarios, this paper proposes a real-time target localization method and system for UAVs based on a 3D electromagnetic map, building upon existing research in electromagnetic mapping. This method utilizes an existing 3D electromagnetic map, extending the localization problem from 2D to 3D. It employs an improved wolf pack algorithm (AGWO) incorporating back-learning to replace the traditional MMSE estimation algorithm for target localization, solving the critical problem of the MMSE algorithm's complexity increasing with dimensionality in 3D scenarios. This achieves low-complexity, high-precision real-time localization, promoting the further application and development of electromagnetic maps in the positioning field.

[0009] This invention is implemented as follows: a real-time target localization method for unmanned aerial vehicles (UAVs) based on a three-dimensional electromagnetic map, comprising the following steps:

[0010] The system calls up a pre-built 3D electromagnetic map of the area to be located. The 3D electromagnetic map is constructed based on static terrain elevation data and the Longley-Rice propagation model and is used to characterize the signal propagation characteristics in a static environment. The system reads the 3D spatial boundary, 3D grid matrix, and theoretical values ​​of the received signal strength per unit transmit power for each UAV receiving node corresponding to each grid node of the area to be located.

[0011] The system acquires real-time observations of the received signal strength from multiple core positioning drones transmitting signals to the target, as well as the real-time spatial position of each core positioning drone.

[0012] Within the search range defined by the three-dimensional spatial boundary, candidate solutions for the target location are generated. For each candidate solution, the theoretical value of the received signal strength per unit transmit power of the corresponding grid node is called, and the theoretical value of the received signal strength corresponding to the candidate solution is calculated based on the target transmit power.

[0013] Candidate solution evaluation results are constructed based on the deviation between the theoretical and observed values ​​of the received signal strength of each core positioning UAV. The candidate solutions for the target location are then iteratively updated based on these evaluation results. During this iterative update process, UAV cooperative control is performed simultaneously: Formation geometric features are calculated based on the real-time spatial positions of each core positioning UAV. These features are compared with a preset threshold. When the geometric features are not greater than the threshold, a tight formation mode is switched, selecting the core positioning UAV with the highest observed received signal strength as the master node, with the remaining core positioning UAVs acting as slave nodes following the master node to accelerate convergence and suppress interference from anomalous single-point observations. When the geometric features are greater than the threshold, an independent formation mode is switched, with each core positioning UAV independently moving towards the current target candidate solution. Simultaneously, when the distance between the auxiliary UAV and the current target candidate solution is less than a preset threshold, a spiral search mode is activated, and the observed received signal strength collected during the spiral search process is fed back into the subsequent candidate solution evaluation, participating in the calculation of the candidate solution evaluation results.

[0014] After the preset iteration termination condition is met, the candidate solution with the best evaluation result is output as the final positioning coordinates of the target.

[0015] This invention also provides a method for real-time target localization of unmanned aerial vehicles (UAVs) based on a three-dimensional electromagnetic map, comprising the following steps:

[0016] Step 1: The three-dimensional electromagnetic map is used.

[0017] 1.1 Electromagnetic Map Foundation: The existing 3D electromagnetic map for complex terrain is directly adopted. This map is constructed based on the Longley-Rice propagation model and real geographic data. Its core construction logic is as follows: the area to be located is discretized into a 3D grid index; a UAV is used as a signal receiving point; all grid points are traversed and simulated as signal transmitting points; the theoretical RSS value corresponding to each grid point is calculated; and finally, a 3D electromagnetic map representing the RSS distribution pattern in 3D space is generated. Simultaneously, the 3D spatial boundary of the area to be located corresponding to this 3D electromagnetic map, i.e., the x-dimensional boundary, is obtained. y-dimensional boundary z-dimensional boundary There is no need to repeat the map building process.

[0018] 1.2 3D Mesh Parameters Adopted: The existing mesh discretization parameters of the 3D electromagnetic map are adopted, and the 3D mesh matrix of the area to be located is... The pre-built, discretely constructed 3D electromagnetic map slice diagram is shown below. Figure 1 As shown, the three-dimensional coordinates of its mesh nodes are: ( ; ; ),in , , These are the grid indices in the x, y, and z dimensions, respectively. , , The total number of grid cells corresponding to the three dimensions is determined by the boundary differences of each dimension and the preset discretization precision of the map. Calculated, i.e. , , This grid matrix is ​​directly used for subsequent positioning calculations.

[0019] 1.3 Collaborative Application of Model Parameters and RSS Theoretical Values: The complete set of parameters for the Longley-Rice propagation model, calibrated during the 3D electromagnetic map construction phase, is used, including transmitter power P_0, transmitter gain G, receiver gain G, operating frequency f, and terrain attenuation coefficient. The terrain attenuation coefficient has been dynamically calibrated based on the terrain undulation of the area to be located, using GMTED2010 elevation data, and does not require reconfiguration. Simultaneously, pre-stored data at each grid node u=v(l,m,n) and at different observation times of each UAV receiver node are accessed from the map. Theoretical value of unit power RSS ( (For UAV indexing), combined with the transmitter power P_0, the theoretical RSS value corresponding to each grid node can be quickly derived without repeated calculations. This 3D electromagnetic map has established a correlation mapping between the 3D coordinates of grid nodes and the theoretical RSS value, and stores complete RSS feature information of the area to be located. It can realize real-time retrieval of relevant data, significantly simplifying the positioning process and ensuring real-time positioning.

[0020] Step 2: Deployment and Signal Acquisition of the Unmanned Aerial Vehicle System

[0021] A positioning system is formed by deploying U-shaped drones, with one drone serving as an auxiliary drone and the rest as core positioning drones; all drones collect RSS observations transmitted by the target in real time. ( For drone indexing, The observation data is transmitted in real time to the positioning processing module for positioning. During the positioning process, the core positioning UAV is deployed at a preset initial position and mainly undertakes the core task of RSS observation required for positioning calculation. The auxiliary UAV is mainly used to help expand the search range and further improve the positioning reliability. The initial deployment position of all UAVs must cover the core area of ​​the area to be positioned to avoid blind spots in the initial signal acquisition. The preset initial position can be flexibly adjusted according to the terrain features and boundary parameters of the three-dimensional electromagnetic map of the area to be positioned.

[0022] The target launch node operates at constant power. Drone continuously emits electromagnetic signals At the observation time Collected RSS observations The following model is satisfied:

[0023]

[0024] in, For drones In the radio map, the grid where the target is located The corresponding theoretical value for a unit power signal; To observe the random disturbance term, which follows a mean of 0 and a variance of , Gaussian distribution The main causes are electromagnetic background noise, UAV sensor measurement errors, and residual errors from multipath effects.

[0025] Step 3: Real-time target localization based on the improved wolf pack algorithm (AGWO)

[0026] 3.1 Population Initialization: The AGWO algorithm population is initialized using an elite reverse learning strategy, with a population size of [missing information]. Individual dimensions of a population correspond to three-dimensional spatial coordinates The value is limited to the three-dimensional grid coverage of the area to be located, ensuring that the population search range is completely matched with the area to be located.

[0027] 3.1.1 Generating the initial random population: generated through a uniform random distribution. The initial individuals constitute the initial random population. Among them, individuals ( The coordinates of each dimension satisfy , , This ensures that each initial individual falls within the area to be located;

[0028] 3.1.2 Generating a reverse population: For any individual in the initial random population Its reverse individual Calculated by the following formula:

[0029]

[0030] The reverse population consists of all reverse individuals. ;

[0031] 3.1.3 Selection of high-quality individuals: Merge the initial random population and the reverse population to obtain a population of size [missing information]. Mixed population Calculate the fitness value of each individual in the mixed population, sort them by fitness value from smallest to largest, and select the optimal one. Individuals constitute the final initial population. This improves the search performance and convergence speed of the initial population;

[0032] 3.2 Fitness Function Design: A fitness function is defined to quantify the search performance of individuals in the population, measuring the overall degree of matching between the observed RSS values ​​and the theoretical RSS values ​​in the 3D electromagnetic map. The smaller the fitness value, the closer the 3D coordinates of the individual are to the actual coordinates of the target. The fitness function is as follows:

[0033]

[0034] in, For the population number The fitness value of each individual For the population number The three-dimensional position vector of each individual. The term represents the mean of the deviation. For discrete constraint terms, These are weighting coefficients used to balance the influence weights of the two items; the default value is... ;

[0035] Deviation mean term Depend on The average absolute deviation between the theoretical and observed RSS values ​​of the core positioning UAV is calculated to reflect the overall matching deviation:

[0036]

[0037] in, ( )for The corresponding grid node at the first Theoretical RSS value of the drone (taken from a 3D electromagnetic map, i.e.) , (Constant transmission power for the target transmitting node). For the first RSS observations of a drone (i.e. );

[0038] Discrete constraint terms calculate The dispersion of the absolute RSS deviation of a single UAV is assessed to suppress signal interference from individual UAVs (such as abnormal observations caused by signal blockage or noise interference), thereby improving the robustness of the fitness function.

[0039]

[0040] in, For standard deviation operator;

[0041] 3.3 Population position update: determined by balance factor (default The position update mode is adjusted to achieve dynamic coordination between global exploration and local development, which avoids the algorithm getting stuck in local optima and ensures convergence accuracy. The specific process is as follows:

[0042] 3.3.1 Update Mode Decision: For each individual in the population, generate Interval random numbers ,like Enter GWO development mode and perform a fine-grained local search; if Enter AO exploration mode to expand the global search range;

[0043] 3.3.2 GWO Development Model: Utilizing Nonlinear Convergence Factors (Calculated by the following formula), the fusion population leadership ( , , Wolf location information drives individuals to perform a fine-grained search for the optimal region, convergence factor The intensity of development is dynamically adjusted based on the number of iterations.

[0044]

[0045] in, This represents the current iteration number. The maximum number of iterations is given; the position update formula is:

[0046]

[0047] in, , , Based on , , The candidate positions for wolf location calculation are integrated with leadership information to improve development accuracy;

[0048] 3.3.3 AO Exploration Mode: Generating Interval random numbers The exploration strategy is to either expand or narrow the exploration area with equal probability.

[0049] Expanding the search range by leveraging the average position of the population to avoid missing optimal areas, the position update formula is:

[0050]

[0051] in This represents the average population location. The numbers are uniformly distributed random numbers;

[0052] Narrowing exploration employs a simplified Levy flight strategy to achieve refined exploration and local optimal escape. The narrowing exploration formula is as follows:

[0053]

[0054] in, The best individual in the current population ( The three-dimensional position vector of the wolf. Let Lévy's flight distribution function be used. (In three-dimensional space), the stochastic nature of Levy flight can effectively escape local optima, and its function is as follows:

[0055]

[0056] s is the step size adjustment constant, with a value of 0.01; This is a shape parameter, with a value of 1.5. These are random numbers distributed according to a standard normal distribution. It is the Gamma function; This is the scale parameter for Levy's flight step length.

[0057] 3.3.4 Population Information Migration: Calculate the new location after the AO exploration mode is updated. The fitness value, if This triggers the "wolf pack migrates to eagle" mechanism, updating the rules as follows: Reorder the fitness values ​​of all individuals in the population and update the leadership ( , , (Wolf) hierarchy, ensuring the population always evolves in the optimal direction;

[0058] Step 4: Unmanned Aerial Vehicle (UAV) Cooperative Control

[0059] This is executed synchronously with population location updates, based on the real-time locations of three core positioning drones. , , Calculate the area of ​​the formation triangle :

[0060]

[0061] Will With formation threshold In comparison, switching between tight formation mode and independent formation mode enables dynamic adjustment of the drone's position, improving the comprehensiveness and accuracy of signal acquisition;

[0062] Tight formation mode ( The core positioning UAV with the highest RSS observation value is selected as the master node. The master node moves to the target candidate position. The remaining two core positioning drones act as slave nodes, synchronously adjusting their positions in sync with the master node. The slave node position update formula is as follows:

[0063] in , , , It is a random unit direction vector;

[0064] Independent formation mode ( ): 3 core positioning drones independently navigate to the target candidate location The formula for the movement step size is:

[0065]

[0066] in For drone speed, For time step;

[0067] The assisted drone is within a preset threshold distance from the candidate target location. At this time, the spiral search mode is activated to expand the search range and avoid missing targets. The spiral search formula is:

[0068]

[0069] in, The angle increment is used to assist the drone in spiral assist, and is used to control the turning angle of each step of the spiral trajectory; The initial radius of the spiral search is the initial distance between the auxiliary UAV and the target candidate position when it begins spiral motion. This is the maximum radius of the spiral search, used to limit the range of the spiral search and avoid searching too far. The spiral radius growth rate is used to control the rate at which the radius increases over time during the spiral search process, thereby regulating the expansion speed of the search range.

[0070] Step 5: Output of local optimal avoidance and localization results:

[0071] After each round of UAV position adjustment, the AGWO iteration calculation is repeated 6 times to obtain 6 candidate target positions. The k-means clustering algorithm was used to filter candidate locations, and the center of each cluster was calculated. and average fitness within clusters The cluster center with the lowest average fitness is selected as the optimal localization result for this round:

[0072]

[0073] When the number of iterations reaches the maximum number of iterations Output the current best individual As the final positioning coordinates of the target This enables real-time positioning.

[0074] Furthermore, in step 1.2, the discrete precision... It can be dynamically adjusted according to the actual positioning accuracy requirements, and the discrete accuracy is... It is positively correlated with positioning accuracy, that is The smaller the value, the finer the mesh, the more mesh nodes, and the higher the positioning accuracy, but the computational cost increases slightly. In practical applications... You can select 0.002 latitude and longitude; for more complex terrains and high-precision scenarios, select a smaller value. Ordinary scenes are selected with a larger size To balance positioning accuracy and computational efficiency.

[0075] Furthermore, in step 3.2, the weighting coefficients... Used to balance the mean term of the deviation With discrete constraint terms The optimal value for this effect was determined through multiple simulation experiments. When complex terrain causes severe signal interference for a single UAV, the value can be appropriately increased. This enhances the suppressive effect of discrete constraint terms.

[0076] Furthermore, in step 3, the population size of the AGWO algorithm... The default value is 30, which represents the maximum number of iterations. The default value is 50; and The settings can be adjusted according to the real-time positioning requirements; when high real-time requirements are needed, the settings can be reduced. and When high positioning accuracy is required, the size can be appropriately increased. and .

[0077] Another objective of this invention is to provide a real-time target localization system for unmanned aerial vehicles (UAVs) based on a three-dimensional electromagnetic map to implement the aforementioned localization method. This system includes:

[0078] The 3D electromagnetic map construction module is used to acquire and preprocess GMTED2010 elevation data, configure Longley-Rice propagation model parameters, complete 3D area gridding, calculate RSS theoretical values, and generate a 3D electromagnetic map containing RSS information of the area to be located, providing prior information support for positioning. This module can run independently and can be repeatedly called after the map is built without repeated construction.

[0079] UAV deployment module: Includes a U-frame positioning UAV, equipped with an RSS signal acquisition module and a data transmission module, used to collect target RSS observations. It transmits data in real time and is equipped with a GPS positioning module to obtain its real-time location coordinates. The data is transmitted to the collaborative control module; one of the drones is equipped with a positioning assistance module and a spiral search control module to perform spiral search tasks and help expand the search range.

[0080] The localization processing module employs an improved wolf pack algorithm (AGWO) to perform population initialization, fitness calculation, and population location updates, combined with the theoretical RSS value from the 3D electromagnetic map. RSS observations collected by drones Completes real-time target localization calculation and outputs candidate target locations. This module has a built-in algorithm parameter configuration interface, which can adjust the population size according to the actual scenario. Maximum number of iterations Parameters;

[0081] Collaborative control module: Used to control the formation switching and position adjustment of the core positioning UAVs, and to receive real-time position information of the core positioning UAVs. , , Calculate the area of ​​the formation triangle , with threshold After comparison, output formation control commands; simultaneously control the activation and operation of the spiral search mode of the auxiliary UAV, receive the position information of the auxiliary UAV and the target candidate position information, output search control commands, and realize UAV cooperative positioning;

[0082] Result output module: Used to output the final positioning coordinates of the target. This module analyzes positioning errors and can simultaneously output auxiliary information such as positioning time and RSS matching deviation, which facilitates effect evaluation in engineering applications.

[0083] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0084] 1. High-precision 3D electromagnetic map construction, adaptable to complex terrain: This invention combines GMTED2010 high-precision elevation data with the Longley-Rice propagation model, fully considering the impact of complex terrain (undulations, vegetation, occlusion, etc.) on radio wave propagation, dynamically adjusting model parameters such as terrain attenuation coefficient, and calculating the theoretical RSS value of grid nodes with small deviation from the actual scene. The constructed 3D electromagnetic map can accurately depict the signal distribution pattern of the area to be located, providing reliable prior information support for high-precision positioning. At the same time, this invention adopts the design of "using the existing map", eliminating the need to repeat the map construction process, greatly simplifying the positioning process, improving positioning efficiency, and solving the problems of cumbersome and repetitive electromagnetic map construction in existing technologies.

[0085] 2. Significantly reduced localization complexity and greatly improved real-time performance: One of the core innovations of this research is the use of an improved wolf pack algorithm (AGWO) to replace the MMSE estimation algorithm in existing two-dimensional localization schemes for three-dimensional target localization; compared to the MMSE algorithm, which requires traversing all grid points in a three-dimensional scene and whose complexity increases exponentially with the dimension, this approach significantly improves localization efficiency. The AGWO algorithm overcomes the shortcomings of other algorithms by using a population search model, eliminating the need to traverse all grid nodes, and its time complexity is O(n log n). It is far less complex than the MMSE algorithm, the computational load is easier to control, and the positioning response time is shortened by 30%. It effectively solves the pain point of excessive complexity of the MMSE algorithm in 3D positioning, can meet the real-time positioning needs of UAVs, and at the same time realizes the optimization and upgrade of the team positioning algorithm.

[0086] 3. Excellent positioning robustness and anti-interference capability: The elite back-learning strategy optimizes population initialization, improving initial population quality and accelerating algorithm convergence. A fitness function integrating the mean deviation term and discrete constraint term effectively suppresses the impact of abnormal observations caused by signal obstruction and noise interference from a single UAV. Combined with UAV cooperative control strategies and local optimum avoidance mechanisms, the algorithm effectively avoids getting trapped in local optima. Positioning error is positively correlated with the grid interval (discrete precision) of the 3D electromagnetic map. In real-world complex terrain scenarios, positioning error can be flexibly adapted by adjusting the discrete precision / interval in the corresponding direction. Its positioning accuracy is on par with existing MMSE 3D positioning schemes, while exhibiting significant technical advantages in positioning robustness, anti-interference capability, and stability of positioning results.

[0087] 4. Wide applicability, flexible deployment, and high engineering value: The system is flexible in deployment, and the grid discretization accuracy can be adjusted according to positioning accuracy requirements. Algorithm parameters (population size) Number of iterations It adapts to terrain scenarios of varying complexity (mountains, hills, etc.); the drone formation mode can be dynamically switched, and the spiral search mode of the auxiliary drone can effectively avoid missing targets and improve positioning reliability; the hardware requirements are simple, the core positioning drone only needs to be equipped with a basic RSS acquisition module, which is inexpensive, easy to be engineered and deployed, and can be widely used in emergency rescue, military reconnaissance, wildlife tracking and other fields.

[0088] 5. Promoting Integrated Applications: This invention proposes a deep fusion method for three-dimensional electromagnetic maps and UAV real-time positioning, which effectively solves the problem of insufficient integration of the two in existing technologies. Furthermore, by utilizing the terrain influence information contained in the electromagnetic map, it solves the core problem of low RSS positioning accuracy and poor real-time performance of UAVs in complex terrain, promoting the coordinated development of electromagnetic map technology and UAV positioning technology. It has significant theoretical and engineering application value.

[0089] The expected benefits and commercial value of the technical solution of this invention after transformation are as follows: The real-time target positioning method and system for UAVs based on three-dimensional electromagnetic maps proposed in this invention have clear commercial transformation prospects and expected benefits. This solution only requires the UAV to be equipped with a basic Received Signal Strength (RSS) acquisition module and a common civilian GPS module, without the need for expensive hardware such as high-precision inertial navigation systems and differential GPS, which can effectively reduce system hardware costs and has good cost-effectiveness. This invention can be widely used in high-value scenarios such as emergency rescue (e.g., mountain search and rescue, personnel positioning at disaster sites), military reconnaissance (e.g., concealed target detection), border patrol, wildlife tracking, and communication interference source positioning. In typical applications such as mountain emergency rescue, this technology can significantly shorten search and rescue positioning time and improve the success rate of search and rescue. In addition, this technology can be packaged into a standardized integrated hardware and software product, and can be exported and licensed to civilian UAV companies, emergency management departments, and defense units, possessing continuous commercial return potential. Therefore, this invention has broad industrialization promotion value and market competitiveness.

[0090] This invention fills a technological gap in the domestic and international industry: it proposes a novel technical solution for real-time UAV positioning assisted by electromagnetic maps. In existing technologies, 3D electromagnetic maps are mainly used for static analysis scenarios such as offline communication link planning, failing to form a closed-loop fusion with real-time UAV positioning. Existing UAV RSS positioning methods often employ free-space propagation models or traditional ergonomic estimation methods, which struggle to accurately describe the nonlinear propagation characteristics of signals under complex terrain conditions, and suffer from high computational complexity and insufficient real-time performance during 3D expansion. This invention integrates a 3D electromagnetic map constructed based on the Longley-Rice propagation model and GMTED2010 elevation data with an improved wolf pack algorithm (AGWO), constructing a full-link positioning mechanism that includes online map retrieval, real-time conversion of unit power RSS theoretical values, calculation of the dual-element fusion fitness function, and UAV collaborative control. The retrieval method of "using the existing map and combining it with real-time conversion of unit power theoretical values" avoids redundant map construction and calculations, reducing positioning latency. A search of domestic and international literature and patent databases revealed no similar technical solution that combines a 3D electromagnetic map based on the Longley-Rice model with the improved wolf pack algorithm (AGWO) and introduces a real-time calling mechanism for unit power theoretical values ​​to achieve real-time 3D RSS positioning of UAVs. Attached Figure Description

[0091] Figure 1 This is a schematic diagram of a 3D electromagnetic map slice;

[0092] Figure 2 This is a schematic diagram of the positioning scene;

[0093] Figure 3This is a flowchart of a real-time target localization system for unmanned aerial vehicles (UAVs) based on a 3D electromagnetic map.

[0094] Figure 4 A schematic diagram of the population update process for improving the Wolf Pack Algorithm (AGWO);

[0095] Figure 5 This is a comparison chart of the positioning errors of various algorithms based on a 3D electromagnetic map.

[0096] Figure 6 This is a comparison chart of the positioning errors of algorithms based on 3D electromagnetic maps and algorithms based on free space propagation models. Figure 7 The search process of four UAVs under a 2D terrain map. Detailed Implementation

[0097] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0098] Example 1: Real-time Static Single-Target Localization in Mountainous Obstruction Scenarios

[0099] First, in a region to be located with significant elevation changes and ridge obstruction, a three-dimensional electromagnetic map is pre-constructed using existing GMTED2010 elevation data and the Longley-Rice propagation model. The map also includes pre-stored three-dimensional spatial boundaries, grid node index relationships, and theoretical values ​​of the received signal strength per unit transmit power for each UAV receiver node. During the localization process, the pre-constructed map data is directly used instead of reconstructing the map.

[0100] Next, three core positioning drones and one auxiliary drone are deployed. The target transmits signals at a constant power, and each core positioning drone synchronously collects and receives signal strength observations and uploads them to the positioning processing module. Then, the positioning processing flow calls the theoretical value of unit power for each candidate position, converts it to the actual theoretical value based on the target's transmission power, and then calculates the mean deviation and discrete constraints between the theoretical value and the observed value of each drone to complete the candidate position optimization. Finally, when the number of iterations reaches the preset maximum value, the current optimal candidate position is output as the final positioning coordinates of the target.

[0101] Example 2: Multi-round iterative collaborative localization within a narrow canyon area

[0102] First, the core positioning drones are initially distributed on both sides of the canyon and at the exit, while auxiliary drones are deployed in the central part of the area. The system calculates the area of ​​the formation triangle formed by the three core positioning drones in real time and compares it with a preset formation threshold.

[0103] Next, when the formation area is large, each core positioning UAV moves independently to the current target candidate position; when the formation area is less than or equal to the threshold, it switches to a tight formation mode, selecting the UAV with the highest received signal strength as the master node, and the remaining UAVs move synchronously with the master node as slave nodes. Then, after each round of UAV position adjustment, each UAV re-collects observations at the new position for use in the next round of candidate solution evaluation.

[0104] Finally, through repeated iterations of "candidate position update - formation adjustment - new observation acquisition - re-evaluation", the process continues until the termination conditions are met.

[0105] Example 3: Using an auxiliary UAV for spiral search and precise positioning during the near-target phase

[0106] First, after coarse localization in the aforementioned scenario, it is determined whether the distance between the assisted UAV and the current candidate target position is less than a preset threshold. Next, when the distance is less than the threshold, the assisted UAV initiates a spiral search trajectory, continuously acquiring received signal strength observations at different radii and azimuth angles according to a set angle increment and radius growth rate. Then, the new observations acquired during the spiral search are transmitted back to the localization processing module in real time, where they are combined with the observations from the core localization UAV to recalculate the mean deviation and discrete constraints, participating in the next round of candidate position iteration. Finally, through fine sampling via the spiral search, the error range in the near-target region is compressed, suppressing local mismatches.

[0107] This invention provides a real-time UAV target localization method based on a 3D electromagnetic map. By combining the Longley-Rice propagation model with GMTED2010 elevation data to construct a 3D electromagnetic map, and replacing the traditional minimum mean square error estimator with an improved wolf pack algorithm (AGWO), it utilizes pre-stored map prior information and real-time received signal strength observations collected by the UAV to achieve high-precision, low-latency 3D real-time localization in complex terrain. The detailed working principle of this method is as follows:

[0108] Step 1: The 3D electromagnetic map is reused. The 3D electromagnetic map under the complex terrain that has been built is directly reused. The 3D electromagnetic map is built by fusing the Longley-Rice propagation model with real geographic data. The 3D spatial boundary of the area to be located, the pre-built 3D grid matrix, and the configured Longley-Rice propagation model parameters corresponding to the map are obtained. The theoretical values ​​of the unit received signal strength RSS at different times at each grid node and each UAV receiving node are called from the map. There is no need to repeat the map building and parameter configuration process.

[0109] Step 2: UAV system deployment and signal acquisition. Deploy a U-shaped UAV system to form a positioning system, with one UAV as an auxiliary UAV and the rest as core positioning UAVs. All UAVs are deployed at preset initial positions and collect the RSS observation values ​​of the electromagnetic signals continuously emitted by the target at constant power in real time. The observation data is transmitted to the positioning processing module in real time. The auxiliary UAV is used to help expand the search range and improve positioning reliability.

[0110] Step 3: Real-time target localization based on the improved wolf pack algorithm (AGWO), including the following sub-steps:

[0111] 3.1 Population Initialization: The AGWO algorithm population is initialized using an elite reverse learning strategy. The population size is N, and the dimensions of the individuals in the population correspond to three-dimensional spatial coordinates, with the values ​​limited to the three-dimensional grid coverage range of the area to be located. An initial random population and a corresponding reverse population are generated, merged, and the best N individuals are selected as the final initial population.

[0112] 3.2 Fitness Function Calculation: Based on the RSS observations collected by the core positioning UAV and the theoretical RSS values ​​retrieved from the 3D electromagnetic map, the fitness value of each individual in the population is calculated; the fitness function integrates the mean deviation term and the discrete constraint term, and balances the influence of the two terms through weighting coefficients;

[0113] 3.3 Population position update: The position update mode is regulated by the balance factor to achieve dynamic coordination between global exploration and local development; random numbers are generated and compared with the balance factor, and the GWO wolf pack algorithm development mode and AO eagle algorithm exploration mode are switched to complete the individual position update; after the AO algorithm exploration, if the new position has better fitness, population information migration is triggered to update the optimal individual and leadership level of the population.

[0114] 3.4 UAV Cooperative Control: Executed synchronously with the population position update, the formation area is calculated based on the real-time position of the core positioning UAVs, compared with the formation threshold, and the close formation mode and independent formation mode are switched to achieve dynamic adjustment of UAV positions; when the distance of the auxiliary UAVs to the target candidate position is less than the preset threshold, the spiral search mode is activated.

[0115] 3.5 Positioning Result Output: After each round of UAV position adjustment, the AGWO iteration is repeated a preset number of times to obtain the target candidate position; the kmeans clustering algorithm is used to filter the candidate positions, and the cluster center with the smallest average fitness is selected as the optimal positioning result for this round; when the number of iterations reaches the maximum number of iterations, the current optimal individual is output as the final positioning coordinates of the target.

[0116] Preferred Implementation Case: Emergency Rescue Positioning Scenario in Complex Mountainous Areas

[0117] This embodiment is applied to an emergency rescue scenario in complex mountainous areas. The target to be located is a signal transmitter carried by a trapped person. The area to be located is Taiping National Forest Park in Xi'an City, Shaanxi Province, which is a mountainous terrain (elevation range 500~3000m, x-dimension represents latitude, range 33.821°~34.019°E, y-dimension represents longitude, range 108.559°~108.757°N). The terrain is undulating and there is vegetation obstruction. Traditional positioning methods have low accuracy and poor real-time performance. The present invention adopts a UAV real-time target positioning method and system based on a three-dimensional electromagnetic map. The specific implementation steps are as follows:

[0118] Step 1: The three-dimensional electromagnetic map is used.

[0119] 1.1 Basic Construction of Electromagnetic Map: The existing 3D electromagnetic map of the mountainous area is called. This map is constructed based on the Longley-Rice propagation model and GMTED2010 elevation data. The core construction logic is as follows: the mountainous area to be located is discretized into a 3D grid index. Using a UAV as a signal receiving point, all grid points are traversed and simulated as signal transmitting points. The theoretical RSS value corresponding to each grid point is calculated to generate a 3D electromagnetic map that represents the RSS distribution law in 3D space. The 3D spatial boundaries of the map are obtained: x-dimensional boundary [33.821°E, 34.019°E], y-dimensional boundary [108.559°N, 108.757°N], z-dimensional boundary [500m, 3000m]. There is no need to rebuild the map repeatedly.

[0120] 1.2 3D Mesh Parameter Discretization: The mesh discretization parameters of this 3D electromagnetic map are used, where the discretization accuracy in the x and y directions is... (Latitude and longitude), with an altitude (z-direction) discrete interval of 10m; 3D mesh matrix It has been pre-built, and the 3D coordinates of the mesh nodes are represented as follows: ,in , , These are the indices of the grid in the x, y, and z dimensions, respectively. The total number of grid cells in the three dimensions is calculated based on the boundary of the region to be located: total number of grid cells in the x-direction. 100 index cells; total number of cells in the y direction 100 index cells; total number of cells in the z-direction With 270 index cells, this grid matrix is ​​directly used for subsequent positioning calculations.

[0121] 1.3 Collaborative Application of Model Parameters and RSS Theoretical Values: The complete set of parameters for the Longley-Rice propagation model, calibrated during the 3D electromagnetic map construction phase, is used, including transmitter power P_0, transmitter gain G, receiver gain G, operating frequency f, and terrain attenuation coefficient. The terrain attenuation coefficient has been dynamically calibrated based on the terrain undulation of the area to be located, using GMTED2010 elevation data, and does not require reconfiguration. Simultaneously, pre-stored data at each grid node u=v(l,m,n) and at different observation times of each UAV receiver node are accessed from the map. Theoretical value of unit power RSS ( (For UAV indexing), combined with the transmitter power P_0, the theoretical RSS value corresponding to each grid node can be quickly derived without repeated calculations. This 3D electromagnetic map has established a correlation mapping between the 3D coordinates of grid nodes and the theoretical RSS value, and stores complete RSS feature information of the area to be located. It can realize real-time retrieval of relevant data, significantly simplifying the positioning process and ensuring real-time positioning.

[0122] Step 2: Deployment and Signal Acquisition of the Unmanned Aerial Vehicle System

[0123] deploy The positioning system is composed of several drones, using small multi-rotor drones (with a flight time of ≥2 hours, equipped with an RSS data acquisition module, GPS positioning module, and data transmission module). Specific positioning scenarios are as follows: Figure 2 As shown, three of them are used as core positioning drones (numbered 1, 2, and 3, i.e. One drone (numbered 4) was used as an auxiliary drone; all positioning drones were deployed at their preset initial positions: Drone 1 was deployed at... Drone 2 is deployed at Drone 3 deployed at Drone 4 deployed in It covers the core area to be located; the core positioning drone collects real-time RSS observations transmitted by the signal transmitter of the trapped personnel. The observation data is transmitted in real time to the ground base via a wireless transmission module (i.e., ); Assist drone 4 is used to help expand the search range.

[0124] Step 3: Real-time target localization based on the improved wolf pack algorithm (AGWO)

[0125] 3.1 Population Initialization: The AGWO algorithm population is initialized using an elite reverse learning strategy. The population size is... Individual dimensions of a population correspond to three-dimensional spatial coordinates The value is limited to the 3D grid coverage area of ​​the region to be located; individuals in the population are uniformly represented as ( ),in , , The first The coordinates of an individual in the x, y, and z dimensions.

[0126] 3.1.1 Generate an initial random population: Generate 30 initial individuals through uniform random distribution to form an initial random population, ensuring that the coordinates of each individual in each dimension meet the boundary requirements of the region to be located, and that all initial individuals fall within the region to be located.

[0127] 3.1.2 Generate a reverse population: For any individual in the initial random population, generate a corresponding reverse individual according to the reverse individual calculation rules, and the reverse population is composed of all reverse individuals;

[0128] 3.1.3 Selection of high-quality individuals: The initial random population and the reverse population are merged to obtain a mixed population of size 60. The fitness value of each individual is calculated by substituting into the fitness function. The individuals are sorted in ascending order of fitness value and the 30 best individuals are selected to form the final initial population, thereby improving the search performance and convergence speed of the initial population.

[0129] 3.2 Fitness Function Calculation: Based on the RSS observations collected by three core positioning UAVs and the theoretical RSS values ​​retrieved from the 3D electromagnetic map, the fitness value of each individual in the population is calculated. The fitness function integrates the mean deviation term and the discrete constraint term. The mean deviation term reflects the overall matching deviation between the RSS observations and the theoretical values, while the discrete constraint term suppresses signal interference from a single UAV. The fusion of these two terms achieves accurate matching and provides a basis for population renewal.

[0130] 3.3 Population Position Update: Setting Balance Factor The dynamic coordination between global exploration and local development is controlled and implemented as follows:

[0131] 3.3.1 Update Mode Decision: For each individual in the population, generate The system generates random numbers within a given interval. Based on the relationship between the random numbers and the balance factor, it switches between GWO development mode (local fine-grained search) and AO exploration mode (global range search).

[0132] 3.3.2 GWO Development Mode: Setting the Maximum Number of Iterations The nonlinear convergence factor changes dynamically with the number of iterations; the fusion population leadership ( , , The system uses wolf location information to calculate candidate locations and update individual locations, enabling fine-grained local search.

[0133] 3.3.3 AO Exploration Mode: Generating Random numbers are generated within an interval, and the search range is expanded or narrowed with equal probability. Expanded search expands the search range by relying on the average position of the population, while narrowed search uses a simplified Levy flight strategy to achieve fine search and local optimal escape. Individual positions are updated according to the corresponding rules.

[0134] 3.3.4 Population Information Migration: After the AO exploration mode is updated, the fitness value of the new location is calculated. If the fitness of the new location is better, the "wolf pack migrates to eagle" mechanism is triggered to update the best individual and leadership level of the population, ensuring that the population evolves in the optimal direction.

[0135] 3.4 UAV Collaborative Control: Executed synchronously with population location updates, based on the real-time positions of the three core positioning UAVs, the area of ​​the formation triangle is calculated, a formation threshold is set, and formation mode switching is achieved.

[0136] In the initial stage, the three core positioning drones were spaced far apart, and the area of ​​the formation triangle was greater than the threshold. Therefore, an independent formation mode was adopted. Each of the three core positioning drones moved independently to the target candidate position. Reasonable drone speed and time step were set, and the position was adjusted according to the corresponding rules to ensure that the movement step was reasonable.

[0137] As the positioning iteration progresses, the three core positioning drones gradually approach the target area. The area of ​​the formation triangle is no greater than the threshold, and the formation mode is switched to close formation. The drone with the highest RSS observation value is selected as the master node. The master node moves towards the target candidate position, and the other two slave nodes follow the master node to move synchronously, so as to avoid the drones being too close or too far apart.

[0138] The system assists the drone in calculating the distance to the candidate target location in real time. When the distance is less than a preset threshold, the spiral search mode is activated. Reasonable spiral search parameters are set, and the spiral search is executed according to the corresponding rules to expand the search range and avoid missing targets.

[0139] 3.5 Local Optimal Avoidance and Localization Result Output: After each round of UAV position adjustment, the AGWO iteration calculation is repeated 6 times to obtain 6 candidate target positions. The k-means clustering algorithm is used to filter the candidate positions, calculate the cluster center and the average fitness within each cluster, and select the cluster center with the smallest average fitness as the optimal localization result for this round to reduce the impact of random errors. The overall improved AGWO algorithm process is as follows: Figure 3 As shown.

[0140] When the number of iterations Reaching the maximum number of iterations Stop iteration and output the current best individual. As the final positioning coordinates of the target ; Calculations show that the final target location coordinates in this embodiment are: It enables real-time positioning in complex mountainous emergency rescue scenarios.

[0141] 3.6 Positioning Accuracy Verification: The final positioning coordinates of the target were compared with the actual position coordinates of the trapped personnel, and the Euclidean distance was calculated as the positioning error. The calculated positioning error was less than 50m, indicating that the positioning accuracy meets the requirements of emergency rescue scenarios. Specific simulation results are as follows: Figure 4 As shown; at the same time, the total positioning time is reduced by 30% compared with the traditional MMSE three-dimensional positioning scheme, fully demonstrating the advantages of low complexity and high real-time performance of the present invention.

[0142] This embodiment verifies the feasibility and superiority of the UAV real-time target positioning method and system based on a three-dimensional electromagnetic map through a real-world application in a complex mountainous environment. Relying on the constructed three-dimensional electromagnetic map and combined with the improved wolf pack algorithm (AGWO), it effectively solves the problems of low RSS positioning accuracy and poor real-time performance in complex terrain. At the same time, through the UAV collaborative control strategy, it improves positioning reliability and can quickly and accurately locate the position of trapped personnel, providing efficient technical support for emergency rescue work.

[0143] Example 1 can be set as "Static Single-Target Real-Time Localization in Mountainous Obstruction Scenarios". In a localization area with significant elevation differences and ridge obstruction, a 3D electromagnetic map is first constructed using existing elevation data and the Longley-Rice propagation model. The map pre-stores the 3D spatial boundaries, grid node index relationships, and theoretical values ​​of the received signal strength per unit transmit power for each UAV receiving node. During localization, the map is not reconstructed; instead, the map data is directly called. Three core localization UAVs and one auxiliary UAV are deployed. The target continuously transmits signals at a constant power, and each core localization UAV synchronously collects and uploads observations. The localization processing flow calls the theoretical value per unit power for each candidate location, then converts it to the actual theoretical value based on the target's transmit power. Subsequently, the mean deviation and discrete constraints are calculated to complete the optimal candidate location selection. This example should focus on clearly describing the map usage, theoretical value calling, observation value matching, termination conditions, and final coordinate output process. This supports the basic closed loop of this application, proving that the scheme does not simply rely on experience search, but uses a pre-built electromagnetic map and real-time observation to form an online positioning mechanism, which has clear feasibility and can be distinguished from traditional schemes that rely solely on RSS ranging or offline fingerprint matching.

[0144] Example 2 can be set as "multi-round iterative collaborative localization within a narrow canyon area". This scenario emphasizes that UAV motion control and localization solution are not two independent modules, but rather work synchronously after each round of candidate position updates. Specifically, it can be written as follows: the core localization UAVs are initially distributed on both sides of the canyon and at the exit position. The system calculates the formation area in real time and compares it with a preset threshold. When the formation area is small, the one with the highest received signal strength is selected as the master node, and the remaining UAVs follow the master node. When the formation area is large, each UAV independently shrinks towards the candidate position. This example highlights that changes in UAV positions will change the observation data in the next round, and the observation data in the next round will in turn affect the candidate position evaluation, thus forming a closed-loop collaboration of "candidate position update - formation adjustment - new observation acquisition - re-evaluation". This description helps to illustrate that the technical contribution of this solution is not just putting flight control and localization algorithms side by side, but using the coupling mechanism of the two to improve convergence efficiency and localization stability in complex terrain, which is sufficient to support the disclosure depth of the core solution in the specification.

[0145] Example 3 can be configured as "assisted UAV spiral search for refined positioning during the near-target phase." After coarse positioning is completed in the aforementioned scenario, when the distance between the assisted UAV and the current candidate position is less than a set threshold, the assisted UAV initiates a spiral search trajectory, continuously collecting received signal strength observations at different radii and azimuth angles, and transmitting the collected results back to the positioning processing flow to re-participate in the calculation of the mean deviation and discrete constraints. This example should clearly state that the spiral search is not a routine cruise maneuver; that is, its triggering condition depends on the coarse positioning result, its feedback object is the evaluation of the next round of candidate positions, and its technical function is to compress the error range in the near-target area and suppress local mismatches. Once this example is implemented, it will support the statement in the specification that "the assisted UAV is not only used to expand the range, but also to refine the candidate position," further enhancing the completeness and verifiability of the solution.

[0146] Example 4 can be used as an extension, set as "layered location in time-varying map correction scenarios." This solution can further add a local correction process: when the deviation between the UAV's observations and the theoretical map values ​​exceeds a set threshold for multiple consecutive rounds, the entire 3D electromagnetic map is not reconstructed; instead, propagation loss correction is applied only to the local grid within the current candidate area before continuing candidate position iteration. The value of this extension lies in advancing "map reuse" to "map reuse plus local adaptive correction." During the defense, this can be proposed as a further limiting alternative technical feature to address examiners' questions about the inaccuracy of the original map due to complex environmental changes. Such reinforcement helps improve inventiveness because it demonstrates that this application does not simply use a static map for a simple table lookup, but rather establishes an online positioning approach that combines map reuse with local correction.

[0147] Example 5 can also be used as an extension, set as "real-time positioning of a three-dimensional electromagnetic map based on multi-band joint observation". Specifically, it can be written as follows: the target transmits detection signals in two or three working frequency bands according to a preset time sequence. The theoretical values ​​of the received signal strength per unit transmit power in each frequency band are pre-stored in the map. During positioning, the theoretical values ​​of multiple frequency bands are called for the same candidate location, and the observation deviations of multiple frequency bands are jointly evaluated. The technical significance of this scheme is that different frequency bands have different sensitivities to terrain occlusion, vegetation attenuation, and multipath effects. Multi-band fusion can reduce the risk of mismatch in a single frequency band. During the defense, this scheme can serve as a direction that can be generalized at a higher level or used to supplement at a lower level, to prove that this application has a deeper design in the utilization of propagation mechanisms, rather than ordinary RSS fitting at a single frequency point.

[0148] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0149] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time target localization method for unmanned aerial vehicles (UAVs) based on a three-dimensional electromagnetic map, characterized in that, Includes the following steps: The system calls up a pre-built 3D electromagnetic map of the area to be located. The 3D electromagnetic map is constructed based on static terrain elevation data and the Longley-Rice propagation model and is used to characterize the signal propagation characteristics in a static environment. The system reads the 3D spatial boundary, 3D grid matrix, and theoretical values ​​of the received signal strength per unit transmit power for each UAV receiving node corresponding to each grid node of the area to be located. The system acquires real-time observations of the received signal strength from multiple core positioning drones transmitting signals to the target, as well as the real-time spatial position of each core positioning drone. Within the search range defined by the three-dimensional spatial boundary, candidate solutions for the target location are generated. For each candidate solution, the theoretical value of the received signal strength per unit transmit power of the corresponding grid node is called, and the theoretical value of the received signal strength corresponding to the candidate solution is calculated based on the target transmit power. Candidate solution evaluation results are constructed based on the deviation between the theoretical and observed values ​​of the received signal strength of each core positioning UAV. The candidate solutions for the target location are then iteratively updated based on these evaluation results. During this iterative update process, UAV cooperative control is executed simultaneously: Formation geometric features are calculated based on the real-time spatial positions of each core positioning UAV. These features are compared with a preset threshold. When the geometric features are not greater than the threshold, a tight formation mode is switched, selecting the core positioning UAV with the highest observed received signal strength as the master node, with the remaining core positioning UAVs acting as slave nodes following the master node to accelerate convergence and suppress interference from anomalous single-point observations. When the geometric features are greater than the threshold, an independent formation mode is switched, with each core positioning UAV independently moving towards the current target candidate solution. Simultaneously, when the distance between the auxiliary UAV and the current target candidate solution is less than a preset threshold, a spiral search mode is activated, and the observed received signal strength collected during the spiral search process is fed back into the subsequent candidate solution evaluation, participating in the calculation of the candidate solution evaluation results. After the preset iteration termination condition is met, the candidate solution with the best evaluation result is output as the final positioning coordinates of the target.

2. The real-time target localization method for UAVs based on a three-dimensional electromagnetic map according to claim 1, characterized in that, The candidate solutions for the target location are initialized using an elite reverse learning approach, which specifically includes: generating an initial random population within the search range; constructing a reverse population corresponding to the initial random population based on the upper and lower bounds of each dimension; merging and evaluating the initial random population and the reverse population; and selecting the top N candidate solutions with the best evaluation results as the initial population.

3. The real-time target localization method for UAVs based on a three-dimensional electromagnetic map according to claim 1, characterized in that, The evaluation result of the candidate solution is determined by the mean deviation term and the discrete constraint term. The mean deviation term is the average value of the absolute deviation between the theoretical value and the observed value of the received signal strength of each core positioning UAV, and the discrete constraint term is the standard deviation of the corresponding absolute deviation of each core positioning UAV.

4. The real-time target localization method for UAVs based on a three-dimensional electromagnetic map according to claim 1, characterized in that, The iterative update of candidate solutions for the target location includes: switching between gray wolf development update mode and eagle exploration update mode according to the balance factor; updating the current candidate solution based on individual information of the population leadership in gray wolf development update mode; updating the current candidate solution using an expanded exploration strategy or a narrowed exploration strategy in eagle exploration update mode; and retaining candidate solutions whose evaluation results are better than those before the update after the candidate solution is updated.

5. The real-time target localization method for UAVs based on a three-dimensional electromagnetic map according to claim 1, characterized in that, During the iterative update of the target position candidate solution based on the candidate solution evaluation results, a UAV cooperative control step is executed simultaneously. The UAV cooperative control step includes: calculating the UAV formation area based on the current round target position candidate solution and the real-time spatial position of each core positioning UAV; comparing the formation area with a preset formation threshold; performing close formation control when the formation area is not greater than the formation threshold; performing independent formation control when the formation area is greater than the formation threshold; and triggering the auxiliary UAV to perform a spiral search when the distance between the auxiliary UAV and the current round target position candidate solution is less than a preset distance threshold.

6. The real-time target localization method for UAVs based on a three-dimensional electromagnetic map according to claim 5, characterized in that, The tight formation control includes: selecting the core positioning UAV with the highest received signal strength as the master node, and the other core positioning UAVs as slave nodes. The slave nodes synchronously adjust their flight positions based on the movement direction of the master node and the candidate solutions for the target position.

7. The real-time target localization method for UAVs based on a three-dimensional electromagnetic map according to claim 5, characterized in that, During the spiral search process, the auxiliary UAV continuously collects and receives signal strength observations and transmits the collection results back to the positioning processing flow to participate in the subsequent candidate solution evaluation result calculation and target position candidate solution iterative update.

8. The real-time target localization method for UAVs based on a three-dimensional electromagnetic map according to claim 1 or 5, characterized in that, A real-time target localization system for unmanned aerial vehicles (UAVs) based on a three-dimensional electromagnetic map includes: The map calling module is used to call a pre-built 3D electromagnetic map and output the 3D spatial boundary of the area to be located, the 3D grid matrix, and the theoretical value of the received signal strength per unit transmit power for each UAV receiving node corresponding to each grid node. The observation and acquisition module is used to acquire the real-time signal strength observation values ​​of the signals transmitted by each core positioning UAV to the target, as well as the real-time spatial position of each core positioning UAV. The positioning and solving module is used to generate candidate solutions for the target position within the search range defined by the three-dimensional spatial boundary, call the theoretical value of the received signal strength per unit transmission power and convert it to the theoretical value of the received signal strength based on the target transmission power, iteratively update the candidate solutions for the target position based on the deviation between the theoretical value of the received signal strength and the observed value of the received signal strength, and output the final positioning coordinates of the target. The collaborative control module is used to perform tight formation control, independent formation control, and auxiliary drone spiral search control based on the current target position candidate solution and the real-time spatial position of each drone.

9. The real-time target positioning system for unmanned aerial vehicles based on a three-dimensional electromagnetic map according to claim 8, characterized in that, The localization solution module includes a population initialization unit, an evaluation calculation unit, and a position update unit. The population initialization unit is used to generate an initial random population and a reverse population. The evaluation calculation unit is used to calculate the mean deviation term and discrete constraint term of each candidate solution. The position update unit is used to switch and update candidate solutions between the gray wolf development update mode and the eagle exploration update mode according to the balance factor.

10. The UAV real-time target positioning system based on a three-dimensional electromagnetic map according to claim 8, characterized in that, The theoretical value of received signal strength per unit transmission power pre-stored in the map calling module is indexed according to the three-dimensional grid nodes, UAV receiving nodes and time information. The positioning solution module converts the theoretical value of received signal strength per unit transmission power obtained by calling according to the target transmission power, and completes the target positioning based on the deviation between the conversion result and the observed value of received signal strength.

Images