Unmanned aerial vehicle safety deployment and early warning method and system for dynamic construction scene

Abstract

This invention discloses a method and system for safe deployment and early warning of unmanned aerial vehicles (UAVs) in dynamic construction scenarios. The method includes: constructing a digital twin model of the construction scenario to achieve real-time perception of dynamic elements; identifying object-machine interaction engineering problems; generating an adaptive learning model for control point deployment with a feedback receiving interface to achieve dynamic optimization of control points; matching a biomimetic model from a biological effects knowledge base according to deployment requirements and replacing its structural features with an object-machine integrated deployment mechanism; using this mechanism as the execution unit, employing a hybrid optimization strategy of deep reinforcement learning and ant colony optimization to achieve multi-objective collaborative planning of the deployment path; quantitatively evaluating the deployment scheme through virtual simulation, establishing a real-time feedback loop between the evaluation results and the adaptive learning model, triggering an early warning and outputting parameter adjustment instructions when indicators exceed limits; and continuously optimizing deployment parameters based on instructions and real-time data collected from the edge to achieve adaptive evolution of the deployment scheme.

Description

Technical Field

[0001] This invention belongs to the field of drone security and construction safety monitoring technology, specifically relating to a drone safety deployment and early warning method and system for dynamic construction scenarios. Background Technology

[0002] With the rapid development of engineering fields such as construction, transportation, and energy, the safety monitoring and management of construction sites are facing increasingly complex challenges. Drones, with their advantages of maneuverability, low cost, and wide coverage, have been widely used in construction site inspection and monitoring, safety deployment, and early warning tasks. However, existing drone safety deployment technologies still have the following technical problems:

[0003] Firstly, in terms of control point deployment, traditional methods rely heavily on manual experience or fixed grid rules, failing to adapt to dynamically changing elements such as obstacles, personnel movement, and machinery movement in the construction scenario. This results in insufficient control coverage or wasted resources. While existing research involves BIM integration with drones or digital twin technology, it primarily focuses on progress monitoring in static scenarios, lacking the ability to perceive and proactively respond to dynamic elements in real time.

[0004] Secondly, in terms of deployment mechanism design, existing UAV deployment mechanisms mostly adopt traditional mechanical structures, and their design ideas are limited to existing experience in the engineering field, making it difficult to break through the path dependence of functional realization. Although biomimetic design has been applied in some fields, it lacks the support of a systematic interdisciplinary knowledge base, resulting in low conversion efficiency between biological prototypes and engineering structures, making it difficult to form a reusable design paradigm.

[0005] Furthermore, in terms of path planning, existing methods mostly employ single optimization algorithms, such as ant colony optimization or genetic algorithms, which struggle to achieve coordinated optimization among multiple objectives, including path length, number of turns, deployment coverage, and dynamic obstacle avoidance capabilities. Simultaneously, the path planning and evaluation feedback processes are disconnected, making it impossible to adjust planning parameters in real time based on evaluation results, thus limiting the dynamic adaptability of the deployment scheme.

[0006] Finally, in terms of assessment and early warning mechanisms, existing technologies mostly adopt single expert assessment or single-dimensional simulation verification, lacking a multi-dimensional quantitative assessment system and real-time feedback closed loop. This makes it difficult to accurately identify the security risks of the deployment scheme, and the early warning response is delayed, making it impossible to achieve continuous iterative optimization of the deployment strategy.

[0007] To address the aforementioned issues, there is an urgent need for a drone safety deployment and early warning method that integrates dynamic perception, interdisciplinary innovation, intelligent planning, and closed-loop optimization, in order to achieve adaptive evolution and synergistic improvement in deployment schemes under construction scenarios. Summary of the Invention

[0008] To address the technical problems existing in current UAV deployment technology for construction scenarios, such as reliance on human experience for control point deployment, limitations of deployment mechanism design to traditional mechanical structures, disconnect between path planning and evaluation feedback, and lack of systematic analysis methods for object-machine interaction, this invention provides a UAV safety deployment and early warning method and system for dynamic construction scenarios. By integrating object-driven design logic, a biological effect knowledge base, and a hybrid optimization strategy, the deployment scheme achieves full-process adaptive evolution.

[0009] In a first aspect, embodiments of this application provide a method for the safe deployment and early warning of unmanned aerial vehicles (UAVs) in dynamic construction scenarios, the method comprising:

[0010] A digital twin model of the construction scene is constructed to realize real-time perception of dynamic elements. Based on virtual simulation, the object-machine interaction engineering problem is explored. Combined with machine learning, an adaptive learning model for control point layout with feedback receiving interface is generated to realize dynamic optimization of control points and output the layout requirements.

[0011] According to the deployment requirements, a biomimetic model is matched from the bio-effects knowledge base, and the biological structural features in the biomimetic model are replaced with the corresponding mechanical structure of the object-machine fusion deployment mechanism to generate the deployment mechanism.

[0012] Using the control mechanism as the execution unit, a hybrid optimization strategy with dynamic parameter adjustment capability is adopted to realize multi-objective collaborative planning of the control path, generating a control scheme that includes the control mechanism and the control path;

[0013] The deployment scheme is quantitatively evaluated through virtual simulation, and a real-time feedback closed loop is established between the evaluation results and the adaptive learning model for the control point deployment. When the indicators exceed the limits, an early warning is triggered, and parameter adjustment instructions are output to the adaptive learning model and the path planning algorithm.

[0014] Based on the instructions and real-time data collected from the edge, the deployment parameters are continuously optimized. After the parameters are updated through the feedback receiving interface, they are sent back to the adaptive learning model and path planning algorithm to achieve adaptive evolution of the deployment scheme.

[0015] Secondly, embodiments of this application provide a drone safety deployment and early warning system for dynamic construction scenarios, applied to the drone safety deployment and early warning method for dynamic construction scenarios as described in the first aspect, the system comprising:

[0016] The digital twin construction module is configured to build a digital twin model of the construction scene to achieve real-time perception of dynamic elements. It explores object-machine interaction engineering problems based on virtual simulation and combines machine learning to generate an adaptive learning model for control point deployment with a feedback receiving interface, so as to realize dynamic optimization of control points and output deployment requirements.

[0017] The biological effect knowledge base module is configured to match a biomimetic model from the biological effect knowledge base according to the deployment requirements, replace the biological structural features in the biomimetic model with the corresponding mechanical structure of the object-machine fusion deployment mechanism, and generate the deployment mechanism.

[0018] The path planning module is configured to use the deployment mechanism as the execution unit and adopt a hybrid optimization strategy with dynamic parameter adjustment capability to realize multi-objective collaborative planning of the deployment path, and generate a deployment scheme that includes the deployment mechanism and the deployment path.

[0019] The quantitative evaluation module is configured to quantitatively evaluate the deployment scheme through virtual simulation, establish a real-time feedback closed loop between the evaluation results and the adaptive learning model for the control point deployment, trigger an early warning when the indicators exceed the limits, and output parameter adjustment instructions to the adaptive learning model and the path planning algorithm.

[0020] The adaptive evolution module is configured to continuously optimize the deployment parameters based on the instructions and real-time data collected from the edge. After updating the parameters through the feedback receiving interface, the module sends the updated parameters back to the adaptive learning model and path planning algorithm to achieve adaptive evolution of the deployment scheme.

[0021] Thirdly, embodiments of this application provide an electronic device, including:

[0022] processor;

[0023] Memory used to store processor-executable instructions;

[0024] The processor is configured to implement the UAV safety deployment and early warning method for dynamic construction scenarios as described in the first aspect when executing the instructions.

[0025] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program that instructs a device to execute the UAV safety deployment and early warning method for dynamic construction scenarios as described in the first aspect.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] This invention constructs a digital twin model of a construction scenario and a mechanism for analyzing the interaction between objects and machines. It uses a matter-element model to standardize the description of engineering problems and combines machine learning and multi-objective priority ranking to achieve a shift from static rules to dynamic adaptation in control point deployment. This solves the technical problems of traditional methods being unable to adapt to dynamic changes in construction scenarios and insufficient deployment coverage. By constructing a biological effect knowledge base and using the Neo4j graph database to build a four-layer architecture storage structure, it replaces the engineering of biological structural features with the mechanical structure of the object-machine integrated deployment mechanism, breaking through the path dependence of traditional mechanical design and providing an interdisciplinary solution for the innovation of deployment mechanisms. Employing a hybrid optimization strategy of deep reinforcement learning and ant colony optimization, the algorithm dynamically adjusts parameters in real time through a policy network, achieving collaborative optimization of the deployment path under multi-objective constraints. Simultaneously, it establishes a real-time feedback closed loop through virtual simulation quantitative evaluation, solving the problems of disconnect between path planning and evaluation feedback and the inability to dynamically adapt. By constructing a quantitative indicator system that includes deployment coverage, path redundancy, point residuals, and risk index, and combining it with a hierarchical early warning mechanism and incremental learning and updating strategy, a complete technical closed loop of planning-execution-evaluation-optimization-evolution is formed, realizing the adaptive evolution of deployment schemes and the synergistic improvement of efficiency. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of a method for safe deployment and early warning of drones for dynamic construction scenarios, provided as an embodiment of this application.

[0029] Figure 2 This application provides an architecture diagram for a drone safety deployment and early warning system for dynamic construction scenarios.

[0030] Figure 3 A schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0032] It should be noted that in the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.

[0033] Based on the embodiments described in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0034] Example 1

[0035] Figure 1 This is a schematic flowchart illustrating a method for safe deployment and early warning of unmanned aerial vehicles (UAVs) in dynamic construction scenarios, provided as an embodiment of this application. Figure 1 As shown, a method for safe deployment and early warning of drones in dynamic construction scenarios includes:

[0036] S1. Construct a digital twin model of the construction scene to achieve real-time perception of dynamic elements. Based on virtual simulation, mine the object-machine interaction engineering problem, and combine machine learning to generate an adaptive learning model for control point deployment with a feedback receiving interface, realizing dynamic optimization of control points and outputting deployment requirements. The function of this step is dynamic perception and deployment requirement generation. Based on the object-machine interaction engineering problem mined by virtual simulation, the object-element model (a formal modeling tool in extensions that combines objects, features, and quantities into ordered triples R=(N, C, V)) is used to standardize the description of the engineering problem, and a two-level object-machine interaction requirement network containing functional elements and carrier elements is constructed. Specifically, this network is constructed as a directed bipartite graph. The first type of node in the graph represents functional requirements (such as collision avoidance), and the second type of node represents carrier requirements (such as tower crane position sensors). When a certain functional requirement needs to be realized through a specific carrier requirement, a directed connection edge is established between the corresponding two types of nodes, and a weight is assigned to the edge according to the association density analysis results. By calculating graph theory indices such as degree centrality and betweenness centrality of each node in the bipartite graph, a two-dimensional criticality judgment based on association density and association degree is achieved.

[0037] Taking a large construction site as an example, the construction area has dynamic elements such as tower crane movement, material transport vehicles shuttling back and forth, and workers moving around, which requires safety control of the drone inspection path.

[0038] First, a digital twin model of the construction site is constructed, integrating BIM data and information collected in real time from multiple sensors to achieve real-time perception of dynamic elements such as tower crane position, vehicle trajectory, and personnel distribution. Virtual simulation is used to model the inspection operations of drones in the construction area. An automated analysis method based on spatiotemporal collision detection is employed to identify the following object-machine interaction engineering problems: interference between tower crane rotation and drone flight paths; obstructed views in material storage areas preventing control points from being visible; and frequent entry and exit of transport vehicles causing temporary occupation of control points. Specifically, the automated analysis method involves real-time calculation of the overlap between the drone's planned path and the axis-aligned bounding boxes (AABBs, i.e., the smallest bounding rectangle aligned with the coordinate axes) of dynamic obstacles (including tower cranes and transport vehicles) in three-dimensional space and time. When the overlap exceeds a preset threshold (e.g., 0.8), the system automatically marks the event in that spatiotemporal coordinate as an object-machine interaction engineering problem and extracts the scene features at that moment (including obstacle type, relative speed, and spatial position) as a structured description of the problem.

[0039] The above engineering problem is described using a matter-element model. Taking the tower crane rotation interference problem as an example, it is expressed as follows: Thing N is the tower crane, feature C is the rotation range, and magnitude v is a fan-shaped area with a radius of 30 meters. Based on the problem description, a two-level object-machine interaction requirement network is constructed: the functional layer requirements include avoiding collisions with the tower crane, maintaining line-of-sight conditions at control points, and avoiding temporarily occupied control points; the carrier layer requirements include tower crane position sensors, material stacking area height measurement devices, and vehicle entry / exit monitoring equipment.

[0040] The degree of mutual influence among various requirements is calculated through a dual-dimensional criticality assessment of correlation density and correlation degree. Correlation density analysis shows that avoiding collisions with tower cranes is strongly correlated with the other five requirements, exhibiting the highest centrality. Correlation degree analysis shows that maintaining line-of-sight conditions at control points has the highest importance weight. Specifically, this embodiment employs a feature importance ranking algorithm based on XGBoost. A training dataset is constructed, where the input features for each sample include: correlation density, correlation degree, historical deployment cost, historical energy consumption, etc.; the output label indicates whether the deployment requirement was adopted in the historical final solution (1 for yes, 0 for no). By training the XGBoost model, the contribution of each input feature to the final decision (i.e., feature importance score) is obtained. The historical deployment data (including point layout plan and actual execution effect) of the past three months is trained and analyzed in combination with machine learning algorithms. Deployment cost index (the labor cost of setting up each control point is about 200 yuan) and energy consumption index (the power consumption of the drone to and from each point) are introduced as economic constraints to perform multi-objective priority ranking. Finally, the three most critical requirements are determined to be: avoid collision with tower crane, maintain line of sight of control points, and avoid temporarily occupied deployment points.

[0041] Based on the sorting results, a dynamic optimization scheme for the location and density of control points is generated: a warning zone is set up at the edge of the tower crane's rotation radius, and the spacing between control points within the zone is reduced from 50 meters to 30 meters; height sensors are added at the edge of the material stacking area to automatically adjust the height of the control points; temporary control points are set up at vehicle entrances and exits, dynamically activated or deactivated based on vehicle entry and exit frequency. The final output is a control requirement list, including 36 fixed control points, 8 dynamically adjustable points, and their corresponding deployment priorities, providing input for subsequent steps.

[0042] The hybrid optimization strategy is a hybrid optimization strategy combining deep reinforcement learning and ant colony optimization. Its dynamic parameter adjustment and multi-objective collaborative planning include the following steps:

[0043] S1.1 The multi-objective optimization function is to minimize the deployment path length, minimize the number of turns, maximize deployment coverage, and maximize dynamic obstacle avoidance capability. In a construction site drone safety deployment task, 36 control points need to be deployed within a 20,000 square meter construction area. The deployment path length should be controlled within 800 meters to reduce drone power consumption; the number of turns should be controlled within 12 to improve flight stability; the deployment coverage rate should reach over 95% to ensure no blind spots; and dynamic obstacle avoidance such as tower crane rotation and vehicle movement is required. These four objectives constitute the multi-objective optimization function, providing optimization direction for subsequent algorithms.

[0044] S1.2. A deep reinforcement learning-based policy network is constructed, taking the state space and action space in the virtual simulation environment as inputs, and dynamically adjusting parameters such as the pheromone evaporation factor, heuristic factor weights, and path selection probability of the ant colony algorithm in real time. In the virtual simulation environment, the state space includes 18 dimensions, such as the current UAV position coordinates, the list of covered control points, the list of uncovered control points, the position and direction of dynamic obstacles, and the remaining battery power. The action space is defined as the target control point number and flight speed level that can be selected for the next step, with a total of 36 discrete action options. The deep reinforcement learning policy network adopts a three-layer fully connected neural network structure. The input layer receives an 18-dimensional state vector, the hidden layer contains 128 neurons, and the output layer corresponds to the Q-value estimation of the 36 actions. The policy network undergoes 5000 iterations of training in the virtual environment, and the reward value is calculated after each iteration. The reward function R is designed as a multi-objective weighted sum, and its specific formula is:

[0045] ,

[0046] in, For the overall reward value, These are the weighting coefficients for the path length target, the number of turns target, the coverage target, and the obstacle avoidance capability target, respectively. These are the scores for path length, number of turns, coverage, and obstacle avoidance. The shorter the current path length, the closer the score is to 1; the closer the current path length is to the maximum allowed value, the closer the score is to 0. The fewer turns you make, the higher your score. Coverage rate is a value between 0 and 1, and is used directly as the score. For example, if the coverage rate is 96%, then... . The lower the collision risk index, the higher the obstacle avoidance score. Weighting coefficients. The initial values ​​can be dynamically set based on the analytic hierarchy process (AHP) or expert experience, and can be set to [0.25, 0.25, 0.25, 0.25]. After training convergence, the policy network can output the parameter combination of the ant colony algorithm in real time according to the current scene state: the pheromone evaporation factor is dynamically adjusted between 0.3 and 0.8, the heuristic factor weight is adaptively adjusted between 1 and 5, and the path selection probability is dynamically corrected according to the real-time state. During the interaction of the hybrid optimization strategy, the policy network outputs a set of dynamically adjusted parameters in real time according to the current state space at each decision step (i.e., before each ant selects the next target control point in the ant colony algorithm).

[0047] S1.3. The dynamically adjusted parameters are input into the ant colony algorithm to guide individual ants in path searching. Iterative training is performed in a virtual simulation environment, using the deployment mechanism as the execution unit to generate a deployment path that meets the multi-objective optimization requirements. The dynamic parameters output by the strategy network are input into the ant colony algorithm. The ant colony size is set to 50 ants. Each ant starts from the starting point and selects the next deployment point based on the weighted probability of pheromone concentration and heuristic factor. When the tower crane in the construction area rotates, the strategy network detects the approach of a dynamic obstacle and immediately increases the pheromone evaporation factor from 0.5 to 0.7, causing the historical path information to decay rapidly and guiding the ants to explore detour routes. At the same time, the weight of the heuristic factor is increased from 3 to 4.5 to enhance the guiding role of the distance factor, causing the ants to prioritize control points far away from the tower crane. After 200 iterations of training, the ant colony algorithm converges to obtain the optimal deployment path: the total path length is 760 meters, the number of turns is 9, the deployment coverage reaches 96%, and it successfully avoids 8 dynamic obstacle events that occur during the simulation. The deployment path uses the object-machine fusion deployment mechanism generated in step 2 as the execution unit. The deployment mechanism completes the deployment actions of 36 control points in sequence according to the path planning results, including the operation sequence of gripping port to grab electrodes, vertical insertion of robotic arm into soil, and automatic release of cable, and finally generates a complete deployment scheme that meets the requirements of multi-objective optimization.

[0048] S2. Based on the deployment requirements, a biomimetic model is matched from the bio-effects knowledge base. The biological structural features in the biomimetic model are engineered and replaced with the corresponding mechanical structure of the object-machine fusion deployment mechanism, generating the deployment mechanism. This step generates the biomimetic mechanism. Based on the deployment requirements output in step 1, a matching biomimetic model is indexed from the pre-built bio-effects knowledge base, biological structural features are extracted, and transformed into the mechanical structure of the object-machine fusion deployment mechanism through an engineering substitution method, generating a deployment mechanism including a clamping port, robotic arm, storage device, and protective structure. This step breaks through the path dependence of traditional mechanical design and provides an interdisciplinary source of innovation for deployment mechanisms.

[0049] Specifically, in this embodiment, the biological effects knowledge base is constructed in the following manner:

[0050] S2.1 Collect biomimetic models containing biological effect names, functional item tags, equipment categories, biological prototype information, biological effect corpus, schematic diagrams, application strategies, and parameter models to form a biological effect corpus.

[0051] Taking the construction of a bio-effect knowledge base for unmanned agricultural equipment as an example, researchers systematically collected biomimetic models from the AskNature bio-strategy platform, biological journal literature, and patent databases. For the bio-effect of earthworm segmental peristalsis, the bio-effect name was recorded as alternating contractions of segmented worm segments, the functional tag was "bristles promoting movement," the equipment category was marked as unmanned agricultural equipment and unmanned construction equipment, the biological prototype information was the earthworm segmental structure and bristle distribution, and the bio-effect corpus was described as earthworms generating wave-like propulsion through alternating segmental contractions, with bristles providing asymmetric friction to achieve unidirectional movement. Simultaneously, structural principle diagrams of this bio-effect were collected, and the application strategy was refined as a multi-stage hydraulic telescopic mechanism + unidirectional locking structure. The parametric model recorded the relationship between propulsion force and friction coefficient. Following the same method, a total of 750 biomimetic models were collected, including desert scorpion surface textures, suckerfish adsorption structures, basking shark gill filtering structures, and bagworm moth multi-layered protective structures, forming a bio-effect corpus covering 11 categories of unmanned engineering equipment.

[0052] S2.2. A four-layer knowledge storage structure is constructed using the Neo4j graph database. The four-layer architecture includes an equipment category layer, a first-level biological function item layer, a second-level biological function item layer, and a biomimetic model layer.

[0053] A four-layer knowledge storage structure is constructed using the Neo4j graph database. The equipment category layer contains 11 nodes, such as unmanned construction equipment, unmanned agricultural equipment, and unmanned mining equipment. Each node records the equipment category name and applicable scenario description. The first-level biological function entry layer contains 30 major functional categories mapped from the biological function classification system, such as movement / stationary, resistance to physical damage, structural mechanics management, and information processing. Each node establishes an applicable relationship with its corresponding equipment category. The second-level biological function entry layer contains 180 specific function entries subdivided from the first-level functions, such as movement in solids, impact resistance, shear / cutting, and perception of tactile and mechanical forces. Each node establishes a membership relationship with a first-level function entry. The biomimetic model layer contains 750 specific biomimetic model nodes. Each node stores attribute information such as the biological effect name, function entry label, biological prototype information, biological effect corpus, schematic diagram storage path, application strategy text, and parameter model description, and establishes an implementation relationship with the second-level function entries. The four-layer architecture forms a complete knowledge graph through the edges between nodes, supporting rapid retrieval from equipment categories down to specific biomimetic models.

[0054] S2.3 Establish a dual-mode retrieval mechanism of linear index and function-oriented index, and push the knowledge base retrieval logic down to the edge computing unit of the UAV to achieve low-latency local indexing and matching.

[0055] A dual-mode retrieval mechanism is constructed to meet the needs of different usage scenarios. The linear index mode is suitable for situations where the equipment type is clearly defined. When a user selects unmanned agricultural equipment, the system sequentially displays primary functional items such as movement / stationary resistance to physical damage and structural mechanics management. If the user selects structural mechanics management, secondary functional items such as shearing / cutting resistance and shock resistance / tremor management are displayed, ultimately locating a biomimetic model, such as a segmented worm body segment alternating contraction model. The function-oriented index mode is suitable for situations where functional requirements are the search entry point. After the user enters "installation mechanism" in the search box, the system calculates semantic similarity using the Sentence-BERT model, automatically matching biological functional items such as movement in a solid, and returning relevant biomimetic models such as earthworms, moles, and snakes.

[0056] Meanwhile, the core retrieval logic of the knowledge base is encapsulated into a lightweight retrieval module and deployed to the edge computing unit of the drone. This module contains only a subset of 200 biomimetic models and an index table related to drone deployment, occupying approximately 200MB of storage space. When the drone performs deployment tasks, it can complete the retrieval and matching locally without communicating with the cloud server. The retrieval response time is reduced from 800 milliseconds for cloud calls to 50 milliseconds for local operation, achieving low-latency local indexing and matching. For example, when the drone detects excessive resistance to electrode insertion in the soil at a construction site, the edge computing unit locally searches for keywords related to low-resistance insertion, quickly matches the earthworm segment peristalsis model, and calls the corresponding robotic arm control parameters in real time to adjust the insertion angle and speed, achieving biomimetic-driven adaptive operation.

[0057] The process of replacing the biological structural features in the biomimetic model with the corresponding mechanical structure of the object-machine fusion control mechanism includes the following steps:

[0058] S2.4 Extract core biomimetic prototype features from the biomimetic model. The core biomimetic prototype features include surface texture topology, multi-level layered protection structure, spiral guide channel structure, and alternating segmental expansion and contraction structure.

[0059] Taking the deployment mechanism design of unmanned farmland electro-disinfection equipment as an example, multiple biomimetic prototype features were retrieved and extracted from the biological effects knowledge base. For the desert scorpion biomimetic model, its surface texture topology was extracted: the scorpion's exoskeleton has periodic microtextures with alternating protrusions and grooves, with protrusions approximately 50 to 150 micrometers high, grooves approximately 100 to 300 micrometers wide, and inclination angles approximately 20 to 35 degrees. This structure effectively reduces soil adhesion and enhances impact resistance. For the bagworm moth biomimetic model, its multi-level layered protective structure was extracted: the cocoon shell consists of 8 to 12 layers of fibers arranged in a cross pattern, with an outer layer of hard chitin, a middle layer of porous buffer layer, and an inner layer of viscoelastic base layer, forming a gradient protection system. For the spiral-shelled animal biomimetic model, its spiral guiding channel structure was extracted: the shell has a spiral downward channel structure, with the pitch gradually decreasing from top to bottom, and a smooth inner wall, which can guide the unidirectional flow of particles and prevent blockage. For the segmented worm biomimetic model, the alternating contraction and expansion structure of its segments is extracted: the earthworm generates wave-like propulsion force through the alternating contraction of longitudinal and circular muscles of the segments, the fluid pressure in the body cavity maintains the stiffness of the segments, and the setae provide asymmetric friction to achieve unidirectional movement.

[0060] S2.5 Establish the functional mapping relationship between biological prototypes and engineering mechanisms, determine the structural amplification factor through the Reynolds number similarity criterion, and realize the parameterized transformation from biological microscopic features to engineering macroscopic dimensions.

[0061] A functional mapping was established for the surface texture topology of desert scorpions: the anti-adhesion function of the scorpion surface microtextures was mapped to the soil adhesion resistance requirement of the electrode plate, and the impact resistance function of the scorpion surface microtextures was mapped to the compressive strength requirement of the electrode plate when inserted into the soil. The structural amplification factor was determined using the Reynolds number similarity criterion. The characteristic scale of the desert scorpion surface microtextures is on the order of 100 micrometers, and the wind speed in the desert environment is approximately 10 to 15 meters per second, while the electrode plate travels at a speed of approximately 2 to 5 meters per second in a farmland environment. Through fluid dynamics similarity calculations, the amplification factor was determined to be 7.5 times, converting the protrusion height to 0.8 mm, the groove width to 1.5 mm, and the structural period to 2.6 mm, while maintaining the tilt angle at 28 degrees. It should be noted that for non-pure fluid dynamics problems involving soil adhesion and impact resistance, this embodiment also incorporates mechanical similarity criteria for comprehensive evaluation. For example, for impact-resistant structures, Cauchy number similarity is used to ensure that the ratio of elastic force to inertial force is consistent between the engineering structure and the biological prototype when subjected to impact; for the process of insertion into the soil, Froude number similarity is referenced to balance the influence of gravity and inertial force. By comprehensively applying multiple similarity criteria, a more precise parametric transformation from biological microscopic features to engineering macroscopic dimensions has been achieved.

[0062] For the multi-layered protective structure of the bagworm moth, a functional mapping was established: the gradient protection function of the bagworm moth cocoon shell is mapped to the protection requirements of the electrode plate when operating in the soil. The outer hard layer corresponds to the wear-resistant layer on the surface of the electrode plate, the middle porous buffer layer corresponds to the stress dispersion structure inside the electrode plate, and the inner viscoelastic base layer corresponds to the buffer pad layer between the electrode plate and the clamping mechanism. Through scalar analysis, the thickness ratio of the cocoon shell layer was applied to the electrode plate design: the thickness of the outer wear-resistant layer was set to 0.5 mm, the thickness of the middle buffer layer was set to 2 mm, and the thickness of the inner buffer pad was set to 3 mm.

[0063] For the helical guide channel structure of helical shell animals, a functional mapping was established: the guiding and anti-clogging function of the shell helical channel is mapped to the particle transport requirements of the particle storage chamber. Through flow channel similarity analysis, it was determined that the helical channel pitch gradually changes from 60 mm at the top to 30 mm at the bottom, and the channel width is set to 80 mm to ensure that graphite particles with a diameter of 25 mm can flow smoothly without clogging.

[0064] For the segmented worm body segment alternating contraction and extension structure, a functional mapping was established: the motion mechanism of earthworm body segment contraction and bristle anchoring was mapped to the insertion and extraction motion requirements of the robotic arm. Through force similarity analysis, the propulsive force generated by earthworm body segment contraction was correlated with the hydraulic thrust of the robotic arm, and the stroke of the vertical mechanism of the robotic arm was determined to be 710 mm, with the push and pull forces set to 2 kN and 3 kN, respectively.

[0065] S2.6. The mapped structural features are replaced by engineering design. The object-machine fusion deployment mechanism includes a clamping port, a vertical mechanism of the robotic arm, a horizontal mechanism of the robotic arm, a storage device, a cable winding and unwinding mechanism, and a protective structure.

[0066] Based on the surface texture characteristics of desert scorpions, an engineering alternative design was developed. Periodic raised and recessed textures were processed on the surface of the electrode plate, with a raised height of 0.8 mm, a groove width of 1.5 mm, and an inclination angle of 28 degrees. The electrode plate was made of 410 stainless steel and had dimensions of 1000 mm long, 250 mm wide, and 2.5 mm thick. The surface texture reduced soil adhesion by approximately 40%.

[0067] Based on the multi-level, layered protection characteristics of bagworm moths, an engineering alternative design was developed, in which a three-layer protective structure was set at the electrode plate clamping port: an outer layer of stainless steel wear-resistant layer, a middle layer of rubber buffer layer, and an inner layer of polyurethane elastic pad, which effectively absorbs the impact load when inserted into the soil and extends the service life of the electrode plate.

[0068] An engineering alternative design was developed based on the characteristics of the spiral guide channel in spiral-shelled animals. A spiral descending channel with a width of 80 mm was set inside the mediating particle storage chamber. The pitch gradually changed from 60 mm at the top to 30 mm at the bottom. The chamber body was made of 304 stainless steel and the inner wall was polished to ensure smooth flow of graphite particles without clogging.

[0069] Based on the alternating extension and retraction characteristics of segmented worm body segments, an engineering alternative design was implemented. The vertical mechanism of the robotic arm adopts a multi-stage hinge linkage and hydraulic drive composite structure. Each of the four linkages is 202.5 mm long, with a hinge angle of 61.2 degrees and a vertical extension stroke of 710 mm. The horizontal mechanism of the robotic arm uses a lead screw and nut drive, with a lead screw length of 1550 mm and a lead of 10 mm, covering the entire storage device in horizontal movement. The gripping port is designed to interlock with the textured surface of the electrode plate, and flexible anti-collision material is added to the inside. The storage device adopts a positive and negative electrode separate packaging design with a 30 mm height difference between the compartments for easy gripping and retrieval. The equipment's safety protection structure adopts a triple protection system of honeycomb sandwich and gradient porous composite. The outer layer is a carbon fiber honeycomb core, the middle layer is a gradient porous aluminum alloy, and the inner layer is a carbon nanotube-reinforced polyurethane composite material. The final result is a complete object-machine integrated deployment and control mechanism, realizing the functions of automatic electrode embedding and retrieval, automatic delivery of mediated particles, and automatic spraying of enhanced liquid.

[0070] S3. Using the control mechanism as the execution unit, a hybrid optimization strategy with dynamic parameter adjustment capability is employed to achieve multi-objective collaborative planning of the control path, generating a control scheme that includes the control mechanism and the control path. The function of this step is intelligent path planning. Using the control mechanism generated in step 2 as the execution unit, a hybrid optimization strategy combining deep reinforcement learning and ant colony optimization is adopted to achieve multi-objective collaborative optimization of path length, number of turns, control coverage, and dynamic obstacle avoidance capability. The policy network constructed by deep reinforcement learning outputs the dynamic adjustment parameters of the ant colony algorithm in real time, guiding path search, and ultimately generating a complete control scheme that includes the action sequence of the control mechanism and the trajectory of the control path.

[0071] Specifically, in this embodiment, the multi-objective collaborative planning generates a deployment scheme through the following steps:

[0072] S3.1. Based on the physical parameters and operational constraints of the control mechanism, a path planning model is established, including standard operation unit division, control mode optimization evaluation, and path key point generation. It should be noted that the method of this invention is not only applicable to safety control in construction scenarios but can also be extended to other fields requiring drone control operations. The following example illustrates a drone-based disinfection operation in a North China Plain farmland. The target farmland area is 30 mu (approximately 20,000 square meters), with flat and regular terrain. First, standard operation units are divided based on the physical parameters of the control mechanism. The electrode plates in the control mechanism are 1000 mm long and 250 mm wide. The electric field distribution is most uniform when the electrode plate spacing is set to 2 meters, and the coverage area for a single operation is limited to 600 square meters by power supply. Accordingly, the 20,000 square meters of farmland is divided into 34 standard operation units, each unit being 30 meters long and 20 meters wide. For each standard operation unit, a path planning model is established, dividing the path within the unit into key points such as entry point, operation path, exit point, and turning path. Using the bottom left corner of the unit as the origin, the longer side as the X-axis, and the wider side as the Y-axis, eight entry and exit points are defined with X-coordinates of 1 meter, 5 meters, 9 meters, 13 meters, 17 meters, 21 meters, 25 meters, and 29 meters, and Y-coordinates alternating between 2 meters and 22 meters, forming a reciprocating operation trajectory. Simultaneously, a turning radius constraint is established, with the minimum turning radius of the UAV set at 2 meters, and the turning path using a semi-circular arc connecting adjacent operation lines.

[0073] S3.2. Three deployment modes—parallel straight line, staggered grid, and ring alternating—are used for evaluation. The uniformity of electric field distribution, path length, and deployment complexity are used as evaluation indicators to select the optimal deployment mode.

[0074] The three deployment modes were evaluated for their effectiveness within the standard operating unit. The parallel linear deployment mode arranges positive and negative electrodes alternately along parallel straight lines, with two rows of electrodes deployed simultaneously in each column. The electrode spacing is 2 meters, resulting in a total of 142 electrodes within the unit, a deployment path length of approximately 160 meters, and a total cable length of approximately 320 meters. The staggered grid deployment mode arranges positive and negative electrodes in a staggered grid pattern, with two rows of electrodes deployed simultaneously in each column, alternating between positive and negative poles. This also results in a total of 142 electrodes, a deployment path length of approximately 160 meters, and a total cable length of approximately 420 meters. The circular alternating deployment mode arranges electrodes in a ring around the unit center, with one row of electrodes deployed in each ring. This results in a total of 142 electrodes, a deployment path length of approximately 402 meters, and a total cable length of approximately 804 meters.

[0075] Using electric field uniformity as the primary evaluation metric, the coefficient of variation of electric field intensity for three modes was calculated through finite element simulation. The coefficient of variation for the parallel linear mode was 0.12, for the staggered grid mode it was 0.10, and for the ring-alternating mode it was 0.18, with the staggered grid mode exhibiting the best uniformity. Using path length as the second evaluation metric, the parallel linear and staggered grid modes both had a path length of 160 meters, while the ring-alternating mode had a path length of 402 meters, showing a clear advantage for the former two. Using deployment complexity as the third evaluation metric, the parallel linear mode allowed for the deployment of two rows of electrodes at a time, simplifying operation; the staggered grid mode also allowed for the deployment of two rows but required alternating positive and negative poles, resulting in slightly higher complexity; the ring-alternating mode could only deploy one row at a time, making it the least efficient. Considering all three metrics, the parallel linear mode performed well in terms of electric field uniformity, path length, and deployment complexity, and had the shortest path length. Therefore, the parallel linear mode was ultimately selected as the optimal deployment mode.

[0076] S3.3. The optimal deployment mode is fused with the deployment path output by the hybrid optimization strategy to generate a complete deployment scheme that includes the action sequence of the deployment mechanism, the trajectory of the deployment path, and the parameters of the deployment mode.

[0077] The deployment path output by the parallel linear deployment mode is fused with the hybrid optimization strategy of deep reinforcement learning and ant colony algorithm. The deployment path output by the hybrid optimization strategy is as follows: starting from the lower left corner of the cell, it travels in a straight line from Y=2 meters to Y=22 meters along the Y direction to complete the deployment of the first column of electrodes; at X=5 meters, it makes a U-turn by turning in a semi-circular arc path with a radius of 2 meters; it then returns in a straight line from Y=22 meters to Y=2 meters along the Y direction to complete the deployment of the second column of electrodes; the above process is repeated until 8 columns of electrodes are deployed, with a total path length of 203.98 meters.

[0078] Based on this path, the following action sequence for the deployment mechanism is generated: At path coordinates X=1 meter, Y=2 meters, the gripping port picks up the positive electrode plate, and the robotic arm vertical mechanism inserts downwards into the soil to a depth of 200 mm. After the electrode is buried, the gripping port releases, and the robotic arm vertical mechanism retracts. The drone travels along the straight path to X=1 meter, Y=22 meters and repeats the above actions to bury the negative electrode plate. In the turning section, the cable reeling mechanism simultaneously releases the electrode cable, with the cable length matching the path length to ensure stable connection. Simultaneously, deployment mode parameters are generated: positive and negative electrode spacing 2 meters, electrode plate surface texture tilt angle 28 degrees, insertion depth 200 mm, voltage 240 volts, current 300 amps, mediated particle delivery rate 6 particles per square meter, and enhanced liquid spraying rate 0.75 liters per cubic meter.

[0079] The final result is a complete deployment scheme that includes the action sequence of the deployment mechanism, the deployment path trajectory, and the deployment mode parameters. It covers 34 standard operation units, with a total of 4,836 deployment electrodes. The operation is expected to take about 12 hours, providing input for subsequent quantitative evaluation.

[0080] S4. Quantitatively evaluate the deployment scheme through virtual simulation, establish a real-time feedback loop between the evaluation results and the adaptive learning model for control point deployment, trigger an early warning when indicators exceed limits, and output parameter adjustment instructions to the adaptive learning model and path planning algorithm. The function of this step is to quantify the evaluation and construct a feedback loop. The complete operational process of the deployment scheme is simulated through virtual simulation, collecting simulation data of quantitative indicators such as deployment coverage, path redundancy rate, point residuals, and risk index, and calculating the overall achievement degree. When any indicator exceeds the limit, a graded early warning signal is triggered, and a real-time feedback loop is established between the evaluation results and the adaptive learning model from step 1, generating parameter adjustment instructions containing indicator deviation data and optimization directions, which are then output to the adaptive learning model and path planning algorithm.

[0081] Specifically, in this embodiment, the step of quantitatively evaluating the deployment scheme through virtual simulation and establishing a real-time feedback closed loop includes the following steps:

[0082] S4.1 Construct a quantitative indicator system that includes control coverage rate, path redundancy rate, point residual, and risk index. Among them, control coverage rate represents the effective coverage of the control area, path redundancy rate represents the efficiency level of path planning, point residual represents the accuracy error of control point deployment, and risk index comprehensively represents the safety risk level of the control scheme.

[0083] Taking a construction site drone safety deployment mission as an example, a four-dimensional quantitative indicator system is constructed. The deployment coverage rate is defined as the ratio of the actual deployment area to the theoretically required deployment area. The theoretical deployment area includes all high-risk locations such as the rotation radius coverage of tower cranes, the edges of material storage areas, and intersections of vehicle access routes, totaling 156 key points, covering an area of ​​approximately 8500 square meters. The path redundancy rate is defined as the percentage of the difference between the actual flight path length and the theoretical shortest path length to the theoretical shortest path length. The theoretical shortest path is calculated to be approximately 3200 meters by traversing the minimum spanning tree of all control points. The point residual is defined as the three-dimensional spatial distance error between the actual control point coordinates and the design coordinates after deployment, obtained through total station measurements. The risk index comprehensively represents the safety risk level of the deployment plan and includes three sub-indicators: dynamic obstacle collision risk, calculated based on the spatial overlap between the tower crane rotation trajectory and the drone flight path; coverage blind spot risk, calculated based on the proportion of high-risk areas not covered by the deployment; and signal obstruction risk, calculated based on the probability of control points losing lock due to building obstruction. The three sub-indicators are weighted to form a comprehensive risk index with a value range of 0 to 1.

[0084] S4.2. Simulate the complete operation process of the deployment scheme through virtual simulation, collect simulation data of each quantitative indicator, and calculate the overall achievement degree of each indicator using the weighted comprehensive method.

[0085] A 3D model of the construction scene was loaded into the digital twin platform. The tower crane was set to rotate once every 90 seconds, transport vehicles randomly entered and exited, and workers were dynamically distributed. The deployment plan was imported into the simulation system to simulate the complete deployment process of 156 control points by drones following the planned path. The simulation was run three times, and the average value was used to collect quantitative index data: the deployment coverage rate reached 94.2%, with 9 high-risk points failing to be deployed due to temporary occupation; the path redundancy rate was 12.6%, and the actual flight path length was 3603 meters; the average point residual was 0.18 meters, and the maximum residual was 0.32 meters; the risk index was 0.23, including collision risk of 0.18, blind spot risk of 0.31, and signal obstruction risk of 0.19. The weighted comprehensive method was used to calculate the overall achievement rate. The weights of each index were determined according to the Analytic Hierarchy Process (AHP, a multi-objective decision-making method that calculates weight coefficients by constructing a judgment matrix): deployment coverage rate weight 0.35, path redundancy rate weight 0.25, point residual weight 0.20, and risk index weight 0.20. After normalizing each indicator and then weighting and summing them, the control coverage rate score was 0.942, the path redundancy rate score was 0.874, the point residual score was 0.820, the risk index score was 0.770, and the overall achievement rate was 0.867.

[0086] S4.3. Compare the overall achievement level with a preset threshold. When any quantitative indicator is lower than the threshold or the risk index exceeds the limit, a graded early warning signal is triggered.

[0087] The preset overall achievement threshold is 0.85. The thresholds for each individual indicator are: coverage rate no less than 93%, path redundancy rate no more than 15%, point residual error no more than 0.20 meters, and risk index no more than 0.25. In this assessment, the overall achievement rate of 0.867 is higher than the threshold, but the average point residual error of 0.18 meters is close to the threshold, while the maximum residual error of 0.32 meters exceeds the threshold. The risk index of 0.23 is lower than the threshold, but the blind spot risk sub-indicator of 0.31 is slightly higher. A level-two warning signal is triggered, with a yellow warning level. The warning content is: "Point residual error exceeds limits; deployment accuracy of 9 points is insufficient; coverage blind spot risk is high; it is recommended to add control points at the edge of the material storage area in the northeast corner."

[0088] S4.4 Establish a real-time feedback closed loop between the evaluation results and the adaptive learning model for control point deployment, generate parameter adjustment instructions containing index deviation data and optimization direction, and output the parameter adjustment instructions to the adaptive learning model and path planning algorithm.

[0089] The evaluation results are used to establish a real-time feedback loop with the adaptive learning model for control point deployment generated in step 1. After receiving the evaluation data, the built-in feedback receiving interface of the adaptive learning model generates parameter adjustment instructions. These instructions include indicator deviation data: the point residual deviation is +0.12 meters, the blind zone risk deviation is +0.06 meters, and the list of uncovered points includes the edge points of stacking areas 3, 5, and 7 in the northeast corner. Optimization directions include: for the issue of excessive point residuals, it is recommended to increase the insertion force control accuracy of the robotic arm's vertical mechanism, adjust the hydraulic system pressure sensor threshold from 2.0 kN to 1.8 kN, and optimize the clamping structure of the gripping port; for the issue of high blind zone risk, it is recommended to add 3 temporary control points in the northeast corner and raise the deployment priority of this area to the highest level. The parameter adjustment command is simultaneously output to both the adaptive learning model and the path planning algorithm: Upon receiving the command, the adaptive learning model updates its training sample set, incorporating the current evaluation data and adjustment plan into its training; the path planning algorithm, upon receiving the command, replans the control path in the northeast corner area. With the addition of control points, the total path length increases by approximately 120 meters, and the number of turns increases by 2, but the control coverage rate improves to 96.8%, and the blind spot risk decreases to 0.22. After three rounds of iterative optimization, the average residual error at the control points decreases to 0.14 meters, the overall achievement rate improves to 0.89, and all indicators meet the threshold requirements, forming a complete evaluation feedback loop.

[0090] S5. Based on the instructions and real-time data collected from the edge, the deployment parameters are continuously optimized. After the parameters are updated through the feedback receiving interface, they are fed back to the adaptive learning model and path planning algorithm to achieve adaptive evolution of the deployment scheme. The function of this step is adaptive evolution and closed-loop iteration. The parameter adjustment instructions output in step 4 and the real-time data collected by the edge computing unit are used as optimization inputs. The optimization input is received through the feedback receiving interface built into the adaptive learning model in step 1, and the model parameters are updated using incremental learning to achieve dynamic optimization of the control point deployment strategy. At the same time, the optimization input is received through the feedback receiving interface built into the path planning algorithm in the hybrid optimization strategy to dynamically adjust the network parameters of the deep reinforcement learning strategy and the pheromone distribution of the ant colony algorithm to achieve online correction of the deployment path. The updated strategy and the corrected path are merged to generate the evolved deployment scheme, and the quantitative evaluation and parameter optimization in step 4 are repeated to form a continuously iterative adaptive evolution closed loop.

[0091] Specifically, in this embodiment, the continuous optimization of deployment parameters based on the instructions and real-time data collected at the edge, to achieve adaptive evolution of the deployment scheme, includes the following steps:

[0092] S5.1 The index deviation data and optimization direction in the parameter adjustment instruction, as well as the real-time data collected by the edge computing unit, are used as optimization inputs. The real-time data includes the actual deployment coverage, path execution deviation, measured residual at the point, and environmental dynamic change parameters.

[0093] During the execution of a drone deployment mission at a construction site, the parameter adjustment instruction generated in step 4 includes the following indicator deviation data: the point residual deviation value is +0.12 meters, the blind zone risk deviation value is +0.06, the list of uncovered points is the edge points of stacking areas No. 3, No. 5, and No. 7 in the northeast corner, and the optimization direction is to add 3 temporary control points in the northeast corner and increase the deployment priority. Meanwhile, the UAV edge computing unit collected real-time field data: the actual deployment coverage rate was 92.8%, a decrease of 1.4 percentage points from the simulation expectation of 94.2%; regarding path execution deviation, the UAV detoured during the third deployment due to the temporary rotation of the tower crane, increasing the actual path length by 85 meters compared to the planned length; the measured residuals at the points were measured on-site by the airborne lidar, with residuals at point 3 in the northeast corner reaching 0.28 meters, point 5 0.31 meters, and point 7 0.26 meters; dynamic environmental parameters included the tower crane's current rotation angle of 135 degrees, material loading and unloading operations underway in the northeast corner stacking area (estimated to last 20 minutes), wind speed of 4.5 meters per second, and light intensity of 32,000 lux. The deviation data in the above parameter adjustment commands, along with the real-time data collected by the edge computing unit, were packaged together as optimization input for subsequent optimization steps.

[0094] S5.2 The optimization input is received through the built-in feedback receiving interface of the adaptive learning model for control point deployment. The training sample set and weight parameters of the model are updated using an incremental learning method to achieve dynamic optimization of the control point deployment strategy.

[0095] After receiving the optimized input, the built-in feedback receiving interface of the adaptive learning model for control point deployment initiates the incremental learning process. The original training sample set of the model contains 1560 control point data points collected over the past six months. Each data point includes scene features, control scheme, and execution effect. The optimized input is added to the training set as new samples: scene features include the construction area, number of tower cranes, distribution of material storage areas, and weather conditions of the day; the control scheme is the currently implemented scheme; the execution effect includes evaluation results such as a mean residual error of 0.18 meters and a coverage rate of 94.2%. Incremental learning adopts a mini-batch gradient descent approach, using new samples to primarily update model parameters, while simultaneously randomly sampling an equal number of old samples from the historical training sample set for joint training. Elastic weights are introduced to consolidate regularization terms to constrain the variation of important weights, thereby maintaining the ability to retain old knowledge while learning new knowledge. The learning rate is set to 0.01, and the batch size is set to 32. After incremental learning, the model weight parameters were dynamically optimized: the control priority weight in the northeast corner area increased from 0.12 to 0.28, the point residual constraint weight increased from 0.15 to 0.23, and the adaptive response weight for temporary occupation increased from 0.08 to 0.19. After the model update, the control point deployment strategy at the edge of the northeast corner stacking area was adjusted from fixed points to dynamically movable points. When material loading and unloading operations are detected, the control points are automatically moved to a temporary reserved position 10 meters away, and then moved back to their original positions after the operation is completed.

[0096] S5.3. The optimization input is received through the feedback receiving interface built into the path planning algorithm in the hybrid optimization strategy. After receiving the optimization input, the parameters of the deep reinforcement learning strategy network and the pheromone distribution of the ant colony algorithm are adjusted simultaneously to achieve online correction of the deployment path. Specifically, the strategy network increases the path selection probability weight in the northeast corner area from 0.15 to 0.32 and expands the path avoidance threshold within the tower crane's rotation radius from 5 meters to 8 meters.

[0097] After receiving the optimization input, the ant colony algorithm dynamically adjusts the pheromone distribution in the northeast corner region. The original pheromones are concentrated on the original planned path. After feedback from the optimization input, pheromone evaporation is applied to the paths of points 3, 5, and 7 where the residual error exceeds the limit. The evaporation factor is set to 0.6, reducing the pheromone concentration on the original path by 40%. Simultaneously, pheromone enhancement is applied to the paths of newly added temporary control points, with an enhancement factor set to 1.5 to attract ants to explore new paths. After 50 iterations, the ant colony algorithm converges to obtain the corrected deployment path, adding three temporary control points in the northeast corner. The total path length is adjusted from the original 3603 meters to 3720 meters, an increase of 117 meters. However, the addition of control points increases the deployment coverage to 96.8%, reduces the blind zone risk to 0.22, and reduces the average residual error to 0.14 meters, achieving multi-objective comprehensive optimization.

[0098] S5.4. Integrate the updated control point deployment strategy with the revised deployment path to generate an evolved deployment scheme, and repeatedly perform the quantitative evaluation and parameter optimization steps to form a continuously iterative adaptive evolutionary closed loop.

[0099] The control point deployment strategy updated after incremental learning is fused with the deployment path corrected by the ant colony algorithm. The control point deployment strategy is updated to use a dynamic, movable point scheme at the edge of the northeast corner of the stacking area, which automatically shifts to a temporary position 10 meters away during material loading and unloading operations. The deployment path is updated to include 3 new temporary control points, with a total path length of 3720 meters and 2 more turns. The fusion generates an evolved deployment scheme, including a total of 159 control points, the action sequence of the deployment mechanism, the deployment path trajectory, and deployment mode parameters.

[0100] The evolved deployment scheme was resubmitted to step 4 for virtual simulation and quantitative evaluation. The second evaluation results showed that the deployment coverage rate increased to 96.8%, the path redundancy rate was controlled at 14.2%, the average point residual error decreased to 0.14 meters, the risk index decreased to 0.20, and the overall achievement rate increased to 0.89. All indicators met the threshold requirements, completing one round of evolution. Subsequently, in actual operations, the drones executed tasks according to the evolved deployment scheme, and the edge computing unit continuously collected real-time data from the site. A local evaluation was performed after each standard operation unit was completed, and a new round of optimization was triggered when deviations were found. After five consecutive rounds of iterative optimization, the average point residual error stabilized at 0.12 meters, the deployment coverage rate stabilized at 97.2%, and the overall achievement rate reached 0.92, forming a continuous iterative closed loop from evaluation to evolution and from evolution back to evaluation, realizing the adaptive evolution of the deployment scheme in dynamic construction scenarios.

[0101] Example 2

[0102] like Figure 2As shown, this application provides an architecture diagram of a drone safety deployment and early warning system for dynamic construction scenarios, applied to the drone safety deployment and early warning system for dynamic construction scenarios as described in Embodiment 1, including:

[0103] The digital twin construction module 210 is configured to build a digital twin model of the construction scene to realize real-time perception of dynamic elements, explore object-machine interaction engineering problems based on virtual simulation, and generate an adaptive learning model for control point deployment with feedback receiving interface by combining machine learning, so as to realize dynamic optimization of control points and output deployment requirements.

[0104] The biological effect knowledge base module 220 is configured to match a biomimetic model from the biological effect knowledge base according to the deployment requirements, replace the biological structural features in the biomimetic model with the corresponding mechanical structure of the object-machine fusion deployment mechanism, and generate the deployment mechanism.

[0105] The path planning module 230 is configured to use the control mechanism as the execution unit and adopt a hybrid optimization strategy with dynamic parameter adjustment capability to realize multi-objective collaborative planning of the control path, and generate a control scheme that includes the control mechanism and the control path.

[0106] The quantitative evaluation module 240 is configured to quantitatively evaluate the deployment scheme through virtual simulation, establish a real-time feedback closed loop between the evaluation result and the adaptive learning model for the control point deployment, trigger an early warning when the index exceeds the limit, and output parameter adjustment instructions to the adaptive learning model and the path planning algorithm.

[0107] The adaptive evolution module 250 is configured to continuously optimize the deployment parameters based on the instructions and real-time data collected from the edge. After updating the parameters through the feedback receiving interface, the module sends the updated parameters back to the adaptive learning model and path planning algorithm to achieve adaptive evolution of the deployment scheme.

[0108] The biological effects knowledge base module includes:

[0109] The corpus storage unit is configured to store a biomimetic model containing biological effect names, functional item tags, equipment categories, biological prototype information, biological effect corpus, schematic diagrams, application strategies, and parameter models.

[0110] The Neo4j graph database unit is configured to construct a four-layer knowledge storage structure, which includes an equipment category layer, a first-level biological function item layer, a second-level biological function item layer, and a biomimetic model layer.

[0111] The retrieval unit is configured to provide a dual-mode retrieval mechanism of linear index and function-oriented index, and the knowledge base retrieval logic is pushed down to the edge computing unit of the UAV.

[0112] Figure 3This is an electronic device provided in one embodiment of this application. For example... Figure 3 As shown, the electronic device includes at least the following components: processor 301 and memory 300, communication interface 303, and bus 302.

[0113] In this embodiment of the application, memory 300 is used to store executable instructions of processor 301, which, when configured to execute instructions, implements the method as described in the first aspect.

[0114] In embodiments of this application, a computer-readable storage medium includes instructions that instruct a device to perform the method as described in the first aspect. For example, the instructions instruct the device to perform... Figure 1 The method is shown in the process steps.

[0115] In one embodiment of this application, the program operating in the electronic device may be a program that controls a central processing unit (CPU) or similar device to achieve the functions of the above-described embodiments of the present invention (a program that enables the computer to function). Information processed by these systems is then temporarily stored in random access memory (RAM) during processing, and subsequently stored in various ROMs such as read-only memory (FlashROM) and hard disk drives (HDDs), and read, corrected, and written by the CPU as needed.

[0116] It should be noted that a portion of the electronic device described above can also be implemented using a computer. In this case, the program for implementing the control function can be recorded on a computer-readable recording medium, and the program recorded on the recording medium can be read into the computer and executed.

[0117] It should be noted that the computer mentioned here refers to a computer built into an electronic device, employing hardware including an operating system and peripheral devices. Furthermore, computer-readable recording media refers to removable media such as floppy disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage systems such as hard drives built into the computer.

[0118] Furthermore, computer-readable recording media can include: media that dynamically stores programs for short periods of time, such as communication lines used when transmitting programs via networks like the Internet or communication lines like telephone lines; and media that store programs for fixed periods of time, such as volatile memory inside a computer that serves as a server or client in this case. In addition, the aforementioned program can be a program used to implement the above-mentioned functions, or it can be a program that can implement the above-mentioned functions by combining them with programs already recorded in the computer.

[0119] Furthermore, the electronic device in the above embodiments can also be implemented as an assembly (system group) composed of multiple systems. Each system constituting the system group can possess some or all of the functions or functional blocks of the electronic device in the above embodiments. As a system group, it is sufficient to have all the functions or functional blocks of the electronic device.

[0120] Those skilled in the art should recognize that the above embodiments are only used to illustrate this application and are not intended to limit this application. Any appropriate changes and variations made to the above embodiments within the essential spirit and scope of this application fall within the scope of protection claimed in this application.

Claims

1. A method for safe deployment and early warning of unmanned aerial vehicles (UAVs) in dynamic construction scenarios, characterized in that, Includes the following steps: A digital twin model of the construction scene is constructed to realize real-time perception of dynamic elements. Based on virtual simulation, the object-machine interaction engineering problem is explored. Combined with machine learning, an adaptive learning model for control point layout with feedback receiving interface is generated to realize dynamic optimization of control points and output layout requirements. According to the deployment requirements, a biomimetic model is matched from the bio-effects knowledge base, and the biological structural features in the biomimetic model are replaced with the corresponding mechanical structure of the object-machine fusion deployment mechanism to generate the deployment mechanism. Using the control mechanism as the execution unit, a hybrid optimization strategy with dynamic parameter adjustment capability is adopted to realize multi-objective collaborative planning of the control path, generating a control scheme that includes the control mechanism and the control path; The deployment scheme is quantitatively evaluated through virtual simulation, and a real-time feedback closed loop is established between the evaluation results and the adaptive learning model for the control point deployment. When the indicators exceed the limits, an early warning is triggered, and parameter adjustment instructions are output to the adaptive learning model and the path planning algorithm. Based on the instructions and real-time data collected from the edge, the deployment parameters are continuously optimized. After the parameters are updated through the feedback receiving interface, they are sent back to the adaptive learning model and path planning algorithm to achieve adaptive evolution of the deployment scheme.

2. The method according to claim 1, characterized in that, The adaptive learning model for control point deployment generates deployment requirements in the following manner: Based on virtual simulation mining, this paper uses the object-element model to standardize the description of the engineering problem and constructs a two-level object-machine interaction requirement network that includes functional elements and carrier elements. By using a dual-dimensional criticality judgment of correlation density and correlation degree, combined with machine learning algorithms to train and analyze historical deployment data, and introducing deployment cost and energy consumption indicators as economic constraints, multi-objective priority ranking is performed to generate a dynamic optimization scheme for control point location and density, and output deployment requirements.

3. The method according to claim 1, characterized in that, The hybrid optimization strategy is a hybrid optimization strategy combining deep reinforcement learning and ant colony optimization. Its dynamic parameter adjustment and multi-objective collaborative planning include the following steps: The multi-objective optimization function is to minimize the deployment path length, minimize the number of turns, maximize the deployment coverage, and maximize the dynamic obstacle avoidance capability. A policy network is constructed using deep reinforcement learning. The network takes the state space and action space in the virtual simulation environment as input and outputs the dynamic adjustment parameters of pheromone evaporation factor, heuristic factor weight and path selection probability of the ant colony algorithm in real time. The dynamically adjusted parameters are input into the ant colony algorithm to guide individual ants to search for paths. Iterative training is performed in a virtual simulation environment, and the deployment mechanism is used as the execution unit to generate a deployment path that meets the requirements of multi-objective optimization.

4. The method according to claim 1, characterized in that, The biological effects knowledge base is constructed in the following ways: Collect biomimetic models containing biological effect names, functional item tags, equipment categories, biological prototype information, biological effect corpus, schematic diagrams, application strategies, and parameter models to form a biological effect corpus; A four-layer knowledge storage structure is constructed using the Neo4j graph database. The four-layer architecture includes an equipment category layer, a first-level biological function item layer, a second-level biological function item layer, and a biomimetic model layer. A dual-mode retrieval mechanism of linear index and function-oriented index is established, and the knowledge base retrieval logic is pushed down to the edge computing unit of the UAV to achieve low-latency local indexing and matching.

5. The method according to claim 1, characterized in that, The process of replacing the biological structural features in the biomimetic model with the corresponding mechanical structure of the object-machine fusion control mechanism includes the following steps: The core biomimetic prototype features are extracted from the biomimetic model. The core biomimetic prototype features include surface texture topology, multi-level layered protection structure, spiral guide channel structure, and alternating segmental expansion and contraction structure. Establish a functional mapping relationship between biological prototypes and engineering mechanisms, determine the structural amplification factor through the Reynolds number similarity criterion, and realize the parameterized transformation from biological microscopic features to engineering macroscopic dimensions; The mapped structural features are then replaced by engineering design to generate a fusion control mechanism for objects and machines that includes a gripping port, a vertical mechanism for the robotic arm, a horizontal mechanism for the robotic arm, a storage device, and a protective structure.

6. The method according to claim 3, characterized in that, The multi-objective collaborative planning generates the deployment scheme through the following steps: Based on the physical parameters and operational constraints of the control mechanism, a path planning model is established, which includes standard operation unit division, control mode excellence evaluation, and path key point generation. Three deployment modes—parallel straight line, staggered grid, and alternating ring—were used to evaluate the effectiveness of the deployment. The uniformity of electric field distribution, path length, and deployment complexity were used as evaluation indicators to select the optimal deployment mode. The optimal deployment mode is fused with the deployment path output by the hybrid optimization strategy to generate a complete deployment scheme that includes the action sequence of the deployment mechanism, the trajectory of the deployment path, and the parameters of the deployment mode.

7. The method according to claim 1, characterized in that, The process of quantitatively evaluating the deployment scheme through virtual simulation and establishing a real-time feedback closed loop specifically includes the following steps: A quantitative indicator system is constructed, which includes control coverage rate, path redundancy rate, point residual and risk index. Among them, control coverage rate represents the effective coverage of the control area, path redundancy rate represents the efficiency level of path planning, point residual represents the accuracy error of control point layout, and risk index comprehensively represents the safety risk level of the control scheme. The complete operation process of the deployment scheme is simulated by virtual simulation, simulation data of each quantitative indicator is collected, and the overall achievement degree of each indicator is calculated by weighted comprehensive method. The overall achievement is compared with a preset threshold. When any quantitative indicator is lower than the threshold or the risk index exceeds the limit, a graded early warning signal is triggered. A real-time feedback loop is established between the evaluation results and the adaptive learning model for control point deployment, generating parameter adjustment instructions that include index deviation data and optimization direction, and outputting the parameter adjustment instructions to the adaptive learning model and path planning algorithm.

8. The method according to claim 1, characterized in that, The continuous optimization of deployment parameters based on the instructions and real-time data collected from the edge, to achieve adaptive evolution of the deployment scheme, specifically includes the following steps: The indicator deviation data and optimization direction in the parameter adjustment instruction, as well as the real-time data collected by the edge computing unit, are used together as optimization inputs. The real-time data includes the actual deployment coverage, path execution deviation, measured residual at the point, and dynamic environmental change parameters. The optimization input is received through the built-in feedback receiving interface of the adaptive learning model for control point deployment. The training sample set and weight parameters of the model are updated using an incremental learning method to achieve dynamic optimization of the control point deployment strategy. The optimization input is received through the feedback receiving interface built into the path planning algorithm in the hybrid optimization strategy, and the parameters of the deep reinforcement learning strategy network and the pheromone distribution of the ant colony algorithm are dynamically adjusted to achieve online correction of the deployment path. The updated control point deployment strategy is integrated with the revised deployment path to generate an evolved deployment scheme. The quantitative evaluation and parameter optimization steps are then repeated to form a continuously iterative adaptive evolutionary closed loop.

9. A drone safety deployment and early warning system for dynamic construction scenarios, applied to the method described in any one of claims 1 to 8, characterized in that, The system includes: The digital twin construction module is configured to build a digital twin model of the construction scene to achieve real-time perception of dynamic elements. It explores object-machine interaction engineering problems based on virtual simulation and combines machine learning to generate an adaptive learning model for control point deployment with a feedback receiving interface, so as to realize dynamic optimization of control points and output deployment requirements. The biological effect knowledge base module is configured to match a biomimetic model from the biological effect knowledge base according to the deployment requirements, replace the biological structural features in the biomimetic model with the corresponding mechanical structure of the object-machine fusion deployment mechanism, and generate the deployment mechanism. The path planning module is configured to use the deployment mechanism as the execution unit and adopt a hybrid optimization strategy with dynamic parameter adjustment capability to realize multi-objective collaborative planning of the deployment path, and generate a deployment scheme that includes the deployment mechanism and the deployment path. The quantitative evaluation module is configured to quantitatively evaluate the deployment scheme through virtual simulation, establish a real-time feedback closed loop between the evaluation results and the adaptive learning model for the control point deployment, trigger an early warning when the indicators exceed the limits, and output parameter adjustment instructions to the adaptive learning model and the path planning algorithm. The adaptive evolution module is configured to continuously optimize the deployment parameters based on the instructions and real-time data collected from the edge. After updating the parameters through the feedback receiving interface, the module sends the updated parameters back to the adaptive learning model and path planning algorithm to achieve adaptive evolution of the deployment scheme.

10. The system according to claim 9, characterized in that, The biological effects knowledge base module includes: The corpus storage unit is configured to store a biomimetic model containing biological effect names, functional item tags, equipment categories, biological prototype information, biological effect corpus, schematic diagrams, application strategies, and parameter models. The Neo4j graph database unit is configured to construct a four-layer knowledge storage structure, which includes an equipment category layer, a first-level biological function item layer, a second-level biological function item layer, and a biomimetic model layer. The retrieval unit is configured to provide a dual-mode retrieval mechanism of linear index and function-oriented index, and the knowledge base retrieval logic is pushed down to the UAV edge computing unit.

Images