Goaf unmanned aerial vehicle inspection system based on monitoring and early warning

Abstract

The invention discloses a goaf unmanned aerial vehicle inspection system based on monitoring and early warning, and the system comprises an observation data collection module which is used for collecting settlement and fracture main variables to form a time stamp observation set; the data fusion module is used for generating a unified monitoring data set by adopting a fractal small-world network coding consensus algorithm; the initial route planning module is used for constructing a route sequence based on a main variable evolution tensor and a risk map; the flight path execution module is used for collecting a flight path and environment data; the fault-tolerant control module triggers a fractional order sliding mode fault-tolerant and degradation mechanism based on the consistency error; and the closed-loop optimization module is used for dynamically updating the main variable tensor and the path cost function and outputting an optimized track and a structured early warning result. The system has high dynamic responsiveness and multi-source risk adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle inspection, and in particular to a goaf area unmanned aerial vehicle inspection system based on monitoring and early warning. Background Art

[0002] With the acceleration of urbanization and the continuous exploitation of mineral resources, surface subsidence, crack expansion and related secondary geological disasters in goaf areas have occurred frequently, posing a serious threat to ground buildings, infrastructure and personnel safety. Therefore, carrying out refined monitoring and early warning of goaf areas has become an important topic in the field of geological disaster prevention and control. Traditional goaf monitoring methods mainly rely on fixed sensor networks (such as GNSS, levels, water pipe inclinometers, etc.) or manual inspection methods. These methods have certain advantages in monitoring accuracy and historical data accumulation, but they generally have problems such as poor deployment flexibility, limited monitoring areas, long data update cycles, and high manual intervention costs. It is difficult to meet the real-time, high-density, and high-efficiency inspection and early warning needs of modern mining areas.

[0003] In recent years, drone remote sensing inspections have been increasingly used in geological hazard monitoring and environmental safety management as a flexible, high-coverage data acquisition method. Their advantage lies in their ability to rapidly cover large areas of complex terrain and acquire high-resolution imagery and spatial data. However, existing drone inspection systems are primarily stand-alone, with path planning relying on rule-based task allocation and static environmental modeling. These systems lack integration with dynamically evolving geological risk information, leading to inefficient task scheduling, redundant flight paths, and missed high-risk areas.

[0004] Furthermore, when it comes to monitoring data processing, traditional methods often perform simple aggregation or statistical analysis of various observational data (such as settlement, cracks, and environmental changes), failing to establish a structured fusion mechanism for multi-source information. Existing fusion algorithms struggle to balance local accuracy and global consistency, especially when data is incomplete and node state changes are inconsistent. While recent research has attempted to introduce graph neural networks and clustering algorithms to model sensor node data, these approaches remain inadequate in characterizing the node state evolution process, spatial disturbance structure, and mission risk distribution, resulting in weak risk identification capabilities.

[0005] At the path planning level, previous research has introduced intelligent algorithms such as reinforcement learning, ant colony algorithms, and particle swarm optimization to optimize drone paths. However, these methods are mostly based on static or single-scale cost functions and fail to consider the "primary variable" driving mechanism in monitoring tasks, namely the differences in risk contributions and path response priorities across timescales at different monitoring points. Existing path optimization methods are unable to adaptively adjust task sequences and trajectory strategies when faced with complex conditions such as multi-target tasks, heterogeneous regional disturbances, and limited inspection resources.

[0006] For fault-tolerant control and mission continuity, traditional UAV control systems generally employ PID, adaptive, or fuzzy control strategies, which only respond to equipment failures or path deviations and lack the ability to warn and handle system-level anomalies caused by "monitoring inconsistencies." During flight, if environmental data deviates from the planned trajectory, the system often cannot promptly identify whether it is due to sensor anomalies, sudden environmental changes, or changes in the UAV's posture, resulting in frequent misjudgments and high mission interruption rates. Existing methods also fail to effectively incorporate dynamic evolutionary control strategies or degradation mechanisms based on risk evolution models, lacking the closed-loop modeling capability for "mission-level" fault-tolerant control.

[0007] Therefore, how to provide a goaf area drone inspection system based on monitoring and early warning is a problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0008] One objective of the present invention is to propose a goaf-based drone inspection system based on monitoring and early warning. This system fully integrates a fractal small-world network coding consensus algorithm, a closed-loop mechanism for principal variable evolution, and a fractional-order sliding mode fault-tolerant control strategy to construct a closed-loop collaborative system consisting of multi-source data acquisition, intelligent trajectory planning, fault-tolerant execution control, and structured early warning output. The system can dynamically respond to the evolutionary characteristics of settlement and cracks, achieving highly consistent data fusion, highly robust trajectory planning, and high-precision early warning identification. It possesses the advantages of strong adaptability, high fault tolerance, and a closed-loop risk response system.

[0009] According to an embodiment of the present invention, a goaf area drone inspection system based on monitoring and early warning includes the following steps:

[0010] The observation data acquisition module is used to collect time series data on surface subsidence and crack evolution characteristics in the goaf. By deploying ground sensor nodes in active subsidence areas and areas with concentrated cracks, it obtains observation data on primary variables such as vertical displacement and crack opening width within each monitoring cycle, and generates an initial observation data set with time tags and node numbers.

[0011] The data fusion module is used to input the initial observation data set into the fractal small-world network coding consensus algorithm for fusion processing, construct a task-sensitive fractal topology structure, perform iterative consensus calculation on the node state vector, and output a unified monitoring data set that meets structural consistency and risk adaptability;

[0012] The initial trajectory planning module is used to construct the main variable evolution tensor and risk nested evolution diagram based on a unified monitoring data set. It combines the main variable trigger frequency with the path cost function construction strategy to generate the initial trajectory planning sequence for the UAV formation and achieve the minimum cost path allocation under multi-objective mission conditions;

[0013] The trajectory execution and data acquisition module is used to control the UAV formation to conduct inspection flights in a timely and orderly manner according to the initial trajectory planning sequence. During the flight, the UAV flight trajectory data, attitude information and environmental status data are collected in real time, and a dual-channel flight-environment data set is generated in a spatiotemporal binding manner.

[0014] The fault-tolerant control module is used to compare the actual flight trajectory with the initial trajectory planning sequence, the current environmental state and the unified monitoring data set, and construct a consistency error function. When the error exceeds the threshold, the control mechanism based on the fractional-order sliding mode fault-tolerant switching and progressive degradation strategy is triggered, and the fault-tolerant control command is output;

[0015] The main variable closed-loop optimization module is used to call the main variable evolution closed-loop mechanism according to the fault-tolerant control instructions and the unified monitoring data set, dynamically update the main variable evolution tensor and path cost function, re-evaluate the path risk and evolution trend of each node, and output the optimized formation trajectory planning scheme and the corresponding structured warning decision results after meeting the convergence conditions;

[0016] The early warning decision output module is used to receive the early warning decision results generated by the main variable closed-loop optimization module, and output structured early warning information including high, medium and low risk level labels, corresponding node numbers and track task adjustment suggestions, and establish associations with historical task execution records for subsequent task maintenance and scheduling optimization.

[0017] Optionally, modules can be connected using the following methods:

[0018] S1. Synchronously collect surface settlement time series and crack evolution characteristic data of the goaf through the ground sensor node network to form the initial observation data set;

[0019] S2, input the initial observation data set into the fractal small-world network coding consensus algorithm for fusion processing, and output a unified monitoring data set;

[0020] S3. Based on the unified monitoring data set, the main variable evolution closed-loop mechanism is used to generate the initial trajectory planning sequence of the UAV formation;

[0021] S4. Control the UAV formation to execute the inspection route according to the initial trajectory planning sequence, and collect formation flight trajectory data and environmental status data in real time;

[0022] S5. Based on the real-time collected data, calculate the consistency error between the flight trajectory and the initial trajectory planning sequence, and the environmental state and the unified monitoring data set; when the consistency error exceeds the preset threshold, generate a warning signal and trigger the fractional-order sliding mode fault-tolerant switching and progressive degradation strategy, and output fault-tolerant control instructions;

[0023] S6. Based on the fault-tolerant control instructions and the unified monitoring data set, the main variable evolution closed-loop mechanism is called to update the trajectory cost function and the main variable evolution tensor state, and the path risk and evolution trend of each node are dynamically re-evaluated. After the convergence conditions are met, the optimized formation trajectory planning scheme and the corresponding warning decision results are output.

[0024] The present invention constructs a multi-source consistency error evaluation mechanism, comprehensively analyzes the path deviation between the real-time flight trajectory of the UAV and the initial trajectory planning, calculates the structural matching degree between the environmental state data and the unified monitoring data set, and introduces a mission risk perception gating function to achieve dynamic monitoring and early warning triggering of the trajectory and environment consistency error. When the error exceeds the set threshold, the system automatically activates the fractional-order sliding mode fault-tolerant controller and the progressive degradation strategy, and generates fault-tolerant control instructions under the premise of ensuring flight safety and mission continuity, thereby improving the robustness and adaptability of the system. This mechanism not only improves the response speed and accuracy of anomaly detection, but also significantly enhances the stability and fault-tolerant control efficiency of UAV inspection tasks under high-risk or sudden environmental changes.

[0025] Optionally, S2 includes the following specific steps:

[0026] S11. Deploy multiple ground sensor nodes, distributed in the active settlement areas and crack concentration areas of the goaf. Each sensor node is equipped with a settlement measurement unit and a crack monitoring unit.

[0027] S12. During each monitoring period, record the surface vertical displacement data and crack opening width data collected by each node, and attach the corresponding time tag and node number;

[0028] S13, the settlement amount and crack parameters collected by each node in the current period are combined into a two-dimensional observation vector, which includes spatial identification and time identification information;

[0029] S14. Summarize all observation vectors in the current period by node number to form a two-dimensional original observation matrix, where rows represent different nodes and columns represent settlement values ​​and crack values ​​of corresponding nodes;

[0030] S15. Perform numerical normalization on the original observation matrix, unify the dimensions, and synchronously calibrate the time tags to generate an initial observation data set.

[0031] The present invention constructs a distributed perception network covering active settlement areas and high-risk fracture areas by deploying ground sensor nodes with settlement and fracture opening monitoring functions in goaf areas. In each monitoring cycle, the system synchronously collects the vertical displacement and fracture width data of each node, constructs a two-dimensional observation vector by combining the node number and time label, and aggregates it into the original observation matrix by spatial number. Subsequently, the dimension unification and time calibration operations are performed to generate an initial observation data set with consistent structure and complete time series. This mechanism ensures the standardized input of multi-source heterogeneous data, provides high-quality basic data for subsequent consensus fusion and risk modeling, greatly improves the system's perception sensitivity and monitoring accuracy of micro-deformation and fracture evolution trends in goaf areas, and realizes efficient capture and early warning response to early signs of geological anomalies.

[0032] Optionally, the fractal small-world network coding consensus algorithm includes the following specific steps:

[0033] Constructing a task-sensitive fractal small-world network topology in the goaf monitoring area includes the following sub-steps:

[0034] Obtain a node set, classify the nodes according to their functional types to form a functional hierarchical structure, classify the surface subsidence monitoring nodes into the inner nested substructure, and classify the drone image recognition nodes into the outer nested substructure;

[0035] Establish a circular local connection relationship based on spatial proximity within each fractal level to generate an adjacent edge set ε local ;

[0036] According to the historical geological risk level r of the area where each node is located j , select remote nodes from different fractal levels to build remote connection edge sets ε remote , where node v i With node v j The probability of long-range connection between ij ;

[0037] At the initialization time t0, for each node v i Construct the observation state vector: where d i (t0) is the node v i The vertical settlement value of the ground collected at time t0, c i (t0) is the crack opening width, s i (t0) is the node structure attribute label, including function type, deployment level and task priority weight;

[0038] Calculate the state covariance matrix Σ of each node i , represents the data distribution structure of the current node in the current observation period, and defines the structural consistency weight between it and the adjacent nodes as:

[0039]

[0040] in,‖·‖ F represents the Frobenius norm, is the set of adjacent nodes of the node;

[0041] Set the main variable dimension to the surface settlement value component d in the node observation state vector i (t k ), calculate its rate of change between two consecutive sampling periods: δ i (t k )=|d i (t k )-d i (t k-1 )|, and set the dynamic threshold function:

[0042] θ(t k )=θ0·(1-β·R(t k ));

[0043] Among them, θ0 is the initial trigger threshold, β is the task risk adjustment coefficient, R(t k ) is the risk level of the system’s current task;

[0044] If the node v i Satisfy δ i (t k )>θ(t k ), the node state update is triggered, and the new state vector is calculated according to the following network coding fusion formula:

[0045]

[0046] in, Represents the state value of the neighbor node in the kth iteration;

[0047] After each round of status update, the consensus termination conditions are determined. If any node meets one of the following conditions, the consensus iteration is terminated:

[0048] The state change amplitude is lower than the set convergence threshold ε, that is,

[0049] The mission risk level reaches or exceeds the warning threshold, that is, R(t k )≥R max ;

[0050] The number of iterations exceeds the maximum limit, that is, k ≥ K max ;

[0051] After consensus is completed, each node outputs the final fusion state Build a unified monitoring dataset:

[0052] The present invention realizes the efficient fusion of ground subsidence and crack evolution data by constructing a task-sensitive fractal small-world network topology. By constructing local adjacency relationships within different functional levels and guiding the generation of remote connection edge sets in combination with historical geological risk levels, the information dissemination efficiency of the monitoring network in complex geological environments is improved. The introduction of a dynamic update mechanism driven by the rate of change of the main variable and a structural consistency weight enables the node state fusion process to maintain spatial correlation while taking into account task risk response, thereby enhancing the convergence and stability of the algorithm under heterogeneous monitoring data. The unified monitoring data set finally constructed not only has structural continuity and temporal integrity, but also improves the expression ability of small deformations and early risk characteristics, significantly enhancing the system's early warning pre-emptiveness and risk identification accuracy.

[0053] Optionally, S3 includes the following specific steps:

[0054] S31, construct the main variable evolution tensor set, and number each monitoring node as v i , extract the main variable observation sequence within T consecutive time periods, including the sedimentation rate sequence, the crack expansion rate sequence and the image structure disturbance degree sequence, and reconstruct it into a three-dimensional tensor: in For node v i The main variable tensor is q, the number of main variable types is q, T is the time window length, and R is the spatial projection scale of the node adjacent area.

[0055] S32. Construct a risk nested evolution diagram based on the main variable tensor of each node in the unified monitoring data set The node set Corresponding to the set of monitoring nodes, the edge weight in the graph is defined as

[0056]

[0057] in, Represents node v i The observation value of the lth dimension of the main variable, K is the sliding window length, w ij (t k ) is the co-evolution edge weight of the main variable;

[0058] S33, using the norm of the state vector of each node in the evolution graph, set the trigger frequency of the local main variable and calculate the node v i The main variable evolution trigger period is:

[0059]

[0060] Where Δt i Represents node v i The local update period, Δt min is the minimum update period, γ is the control adjustment factor, h i (t k ) is the node state vector;

[0061] S34. Extract the node set that triggers the update, generate the track cost distribution map R(x,y) based on the risk projection of the main variable tensor in two-dimensional space, and construct the path cost function of each drone as

[0062]

[0063] in, represents the point sequence of the UAV numbered j in the initial path planning, λ is the risk adjustment coefficient, Cost j is the corresponding path cost;

[0064] S35. Optimize the allocation of paths based on the minimum cost principle, and record the minimum cost path set as The path set is the initial trajectory planning sequence of the UAV formation.

[0065] The present invention realizes refined path modeling for complex risk areas in goafs by constructing a main variable evolution tensor and a multi-dimensional risk nested evolution graph. The system first extracts multi-source main variables such as settlement rate, crack expansion rate and image disturbance degree, and constructs them into a three-dimensional tensor, thereby realizing risk evolution modeling with spatiotemporal continuity; then, a weighted graph structure is constructed based on the co-evolution relationship of the main variables to effectively capture the dynamic correlation and local trigger frequency characteristics between nodes. The path cost function is constructed through risk projection, and a dynamic cost adjustment mechanism is introduced to perform task-sensitive optimization configuration on the UAV track, realizing adaptive allocation of task paths and inspection priority control. This mechanism significantly improves the robustness and risk response capability of path planning, ensuring efficient avoidance and coverage balance of the formation in high-risk areas.

[0066] Optionally, S4 includes the following specific steps:

[0067] S41. Based on the initial trajectory planning sequence, the trajectory point sets of each UAV are issued in sequence to determine the three-dimensional spatial route that each UAV needs to execute, clarify the position parameters of the starting point, target point, and intermediate control points, and set the corresponding flight scheduling order and time interval;

[0068] S42. Control each UAV to enter the target flight segment according to the planned sequence. Based on the time-triggered control strategy, maintain the speed and attitude consistent with the track point set at continuous time, perform spatial position updates at fixed intervals, and record the current flight position, velocity vector, and attitude angle parameters in real time.

[0069] S43. Synchronously collect environmental status data corresponding to the current flight position at each moment, including air temperature, humidity, wind speed, particle concentration, electromagnetic disturbance, remaining battery ratio, and image light intensity information. The environmental status data is structured and stored in a manner that binds time tags and spatial location information;

[0070] S44, performing time synchronization fusion on the flight status data and environmental status data of each UAV during the execution of the current track segment, generating a flight-environment dual-channel time series dataset of the continuous inspection segment, and marking the current mission segment number and track segment type label;

[0071] S45. The flight trajectory data and environmental status data collected by all UAVs in the formation during the current inspection cycle are aggregated to generate a complete inspection execution data set.

[0072] The present invention constructs a flight-environment dual-channel data acquisition mechanism for formation inspection tasks through real-time scheduling control based on the initial track planning sequence, which significantly enhances the task execution accuracy and data relevance in goaf monitoring tasks. In the specific implementation, the system sets a clear three-dimensional track point set and timing scheduling rules for each UAV, and accurately controls the flight speed and attitude based on the time triggering strategy, so that the inspection process has high stability. During the flight, multi-source environmental status data such as air parameters, wind speed disturbances, image light intensity, battery power, etc. are collected at the same time, and structured and fused with the flight status data to form a high-frequency synchronized dual-channel time series data sequence. Through this mechanism, the correspondence accuracy between the flight path and the environmental information is effectively improved, providing high-reliability basic data support for subsequent consistency error judgment and fault-tolerant control, and also significantly improving the completeness and traceability of the inspection data.

[0073] Optionally, S5 includes the following specific steps:

[0074] S51. Based on the inspection execution data set, obtain the current time t of the flight mission k The actual flight trajectory collection The initial trajectory planning sequence of the generated UAV formation And construct the trajectory consistency error function based on the Euclidean distance:

[0075]

[0076] Where N is the number of drones, pa,i (t k ) and p r,i (t k ) represent the actual and reference positions of the i-th UAV respectively;

[0077] S52, synchronously collect task environment state data vector E(t k ), extract the reference environment state vector E at the same moment from the unified monitoring data set * (t k ), construct the state consistency error function: ∈ e (t k )=‖E(t k )-E * (t k )‖2;

[0078] S53, the trajectory error ∈ p (t k ) and state error ∈ e (t k ) is fused into the total consistency error:

[0079] ∈(t k )=γ1·∈ p (t k )+γ2·∈ e (t k );

[0080] Wherein, γ1 and γ2 are preset weighting coefficients, satisfying γ1+γ2=1;

[0081] S54, when ∈(t k )>∈ th When , the warning signal is output, triggering the degradation switching and control mode switching of the main sliding mode controller in the task node set v according to the fractional order sliding mode fault tolerance switching and progressive degradation strategy, and dynamically adjusting the control input vector u(t k );

[0082] S55, output adjusted fault-tolerant control instruction set And update the execution status in the controller lineage structure according to the current degradation path node.

[0083] The present invention constructs a fractional-order sliding mode fault-tolerant switching and degradation control framework for real-time anomaly identification and response during the inspection process by integrating the dual consistency error judgment mechanism of flight trajectory and environmental state. In actual flight, the system dynamically evaluates the deviation between the actual trajectory of the UAV and the initial trajectory based on the Euclidean distance, and simultaneously compares the difference between the current environmental state and the reference state in the unified monitoring data set to calculate the total consistency error value. When the error exceeds the preset threshold, the system immediately issues an early warning, and drives the degradation switching logic through the main sliding mode controller to automatically determine the control state of the task node, and guide the reconstruction of the control mode and the re-estimation of the input vector within the sliding mode surface. At the same time, the execution state in the controller pedigree structure is dynamically adjusted with the degradation path to ensure that the system can maintain stable operation and continuous execution of tasks in extreme or unstable environments. This mechanism significantly improves the fault tolerance and intelligent self-adaptation level of the inspection system, and effectively solves the problem that traditional path planning cannot respond to environmental disturbances in real time.

[0084] Optionally, the fractional-order sliding mode fault-tolerant switching and progressive degradation strategy includes the following specific steps:

[0085] Build a collection of task nodes At each node Collect the main variable evolution tensor And combined with the risk nested evolution diagram to generate the local risk intensity value ρ i (t), define the risk perception function based on task disturbance and environmental uncertainty:

[0086]

[0087] Among them, δ0 is the initial scale of the sliding mode domain, β is the task risk adjustment coefficient, Represents the gradient norm of the tensor in the time axis direction;

[0088] At each node v i Construct a fractional sliding surface:

[0089]

[0090] in, represents the Caputo fractional derivative, order α i (t) Dynamically adjust with risk perception function; e i (t) is the difference between the target track and the actual track; λ i (t) is the environmental state sensitive adjustment gain;

[0091] Aggregate the sliding mode control surfaces of each node to form a regional control law set The sliding mode approach domain is defined as: And according to the ranking results of risk perception functions in each region, a weighted combination of regional control laws is performed:

[0092] Constructing a degenerate lineage tree Each node M j represents a control model with reduced precision, and the degradation path is determined by the following objective function:

[0093]

[0094] Among them, C j (t) is the task cost, F j (t) is the controller's remaining execution capability indicator, R j (t) is the current environment robustness score, η1, η2, η3 are normalized weights;

[0095] At each moment, according to the currently selected degradation model M * (t) Switch the control law set and replace the original sliding mode surface with the weak sliding mode structure in the corresponding degenerate node to perform continuous degradation control;

[0096] Perform a backtracking update operation on the degradation sequence. When the task error is lower than the tolerance range and the risk function continuously decreases for more than the set period, it automatically performs an upward transfer back to the previous layer node M j-1 , and reconstruct the sliding mode controller parameters.

[0097] The proposed fractional-order sliding mode fault-tolerant switching and progressive degradation strategy achieves a robust control mechanism for multi-source disturbances and environmental uncertainties by constructing a risk-aware function and adaptive sliding surface for task nodes. The system first calculates the local risk intensity based on the main variable evolution tensor and evolution graph, and dynamically adjusts the sliding surface order and adjustment gain accordingly, thereby improving control flexibility and response accuracy in extreme environments. Furthermore, the control strategy incorporates a regional sliding mode control law aggregation mechanism, constructing a multi-scale control combination through risk prioritization to achieve a focused response to high-risk areas. During the control degradation process, the system constructs a degradation tree and adaptively selects degradation paths based on task cost, control capability, and environmental score, achieving a smooth transition to a lower-order control model to maintain stable system operation. When the task risk decreases and error recovery conditions are met, the system performs a backtracking update, gradually returning to a high-precision control state, exhibiting excellent self-recovery and global stability. This mechanism significantly improves the fault-tolerant stability and control robustness of the UAV system in dynamic inspection missions in goaf areas.

[0098] Optionally, S6 includes the following specific steps:

[0099] S61, set the fault-tolerant control instruction set With the unified monitoring dataset as the joint input, for each trigger node in the unified monitoring dataset, the main variable evolution closed-loop mechanism is called to update the main variable tensor and calculate the risk intensity. The risk intensity is defined as the product of the norm increment of the node main variable tensor in the last two cycles and the current path error weight.

[0100] S62、The risk intensity R i The node path cost C is obtained by weighted summing the node environmental disturbance index, power threshold and historical track deviation within the mission window. i ;

[0101] S63. For each UAV, the node path costs of all path points are weighted and accumulated to obtain the total cost of the candidate track. The candidate path point sequence is then subjected to the following steps: adjusting the path point order to minimize the total cost; inserting risk avoidance points into high-risk sections; and deleting invalid redundant sections.

[0102] S64, detect the change amplitude of the main variable tensor in multiple consecutive scheduling cycles Total path cost and node risk sequence The iteration is terminated when any of the following conditions is met: ΔX<ε x , ΔC<ε c , ΔR<ε r , reaching the maximum number of iterations K max , where ε x is the convergence threshold of the main variable tensor, ε c is the convergence threshold of the total path cost, ε r is the convergence threshold of the node risk sequence;

[0103] S65, output the optimized formation trajectory planning scheme at the current moment; at the same time, based on the risk intensity R of each node i The evolution trend of the main variables at the time of convergence termination generates a structured warning decision result, including: node risk level label, defined as: high R i ≥T h , Middle T l ≤R i <T h , low R i <T l , Warning node number list, formation task adjustment suggestions, where T h ,T l is the risk level threshold.

[0104] The present invention dynamically updates the main variable evolution tensor and calculates the node risk intensity by integrating fault-tolerant control instructions with a unified monitoring data set, and comprehensively evaluates the path cost in combination with environmental disturbances, power thresholds, and historical track deviations. During the path reconstruction process, risk avoidance, path optimization, and redundancy elimination operations are performed to minimize the total cost. In multiple rounds of iterations, by monitoring the amplitude of the main variable changes, the total path cost, and the risk sequence, it is determined whether to terminate the optimization based on the set convergence threshold. Finally, the optimized formation trajectory planning scheme and structured warning results are output, including risk level labels, warning node numbers, and task adjustment suggestions, to improve the system's trajectory adaptability and warning accuracy in complex environments.

[0105] The beneficial effects of the present invention are:

[0106] This method introduces a closed-loop mechanism for the evolution of primary variables. It uniformly models the settlement rate, crack propagation rate, and image disturbance degree of each monitoring node within different time windows as a three-dimensional evolution tensor. It then constructs a risk-nested evolution graph. This strategy utilizes the local primary variable trigger period and the node path cost function to construct a trajectory optimization strategy, enabling a dynamic response to the evolutionary trends of the primary variables during path allocation. This mechanism significantly outperforms existing methods that select paths based solely on static costs or distance metrics, offering greater mission adaptability and risk coverage.

[0107] During the trajectory execution phase, the present invention constructs a dual-channel flight-environment dataset to collect the drone's position, velocity vector, and multi-dimensional environmental state in real time, and establishes a trajectory-state consistency error function. When the error exceeds a preset threshold, the system automatically triggers a fractional-order sliding mode fault-tolerant switching mechanism and guides the controller to switch to a lower-order control model along the degradation spectrum path. This achieves flexible response to complex factors such as external disturbances, energy degradation, and path deviation, significantly enhancing the robustness and sustainability of mission execution.

[0108] Unlike traditional early warning mechanisms, which structurally separate risk identification from path control, this invention integrates fault-tolerant control results with a closed-loop mechanism for primary variable evolution. It iteratively optimizes the trajectory structure using dynamically updated risk intensity and node cost functions, and performs joint convergence judgments on the primary variable tensor states, total cost functions, and node risk sequences over multiple scheduling cycles. The system's final output includes an optimized formation trajectory planning scheme and structured early warning decision results. It features clear node risk level labels, mission adjustment suggestions, and the ability to correlate historical records, establishing a complete "monitoring-identification-planning-fault-tolerance-early warning" closed-loop path.

[0109] Through the synergistic effects of dynamic adjustment of the sliding surface fractional derivative, node-level controller degradation combination, regional risk perception aggregation and path reconstruction strategy, the present invention enables the system to achieve dynamic task adjustment and precise early warning implementation in complex scenarios such as sudden settlement crack expansion, energy shortage, and path instability. It has high engineering adaptability and deployment feasibility, and is particularly suitable for unmanned monitoring and emergency response scenarios in areas prone to geological disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0110] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0111] Figure 1 This is a flow chart of a goaf area drone inspection system based on monitoring and early warning proposed by the present invention;

[0112] Figure 2 This is a schematic diagram of the fractal small-world network coding consensus algorithm proposed in the present invention;

[0113] Figure 3 This is the execution logic diagram of the fractional-order sliding mode fault-tolerant switching and progressive degradation strategy proposed in the present invention. DETAILED DESCRIPTION

[0114] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0115] refer to Figure 1-3 , a goaf area drone inspection system based on monitoring and early warning, including the following steps:

[0116] The observation data acquisition module is used to collect time series data on surface subsidence and crack evolution characteristics in the goaf. By deploying ground sensor nodes in active subsidence areas and areas with concentrated cracks, it obtains observation data on primary variables such as vertical displacement and crack opening width within each monitoring cycle, and generates an initial observation data set with time tags and node numbers.

[0117] The data fusion module is used to input the initial observation data set into the fractal small-world network coding consensus algorithm for fusion processing, construct a task-sensitive fractal topology structure, perform iterative consensus calculation on the node state vector, and output a unified monitoring data set that meets structural consistency and risk adaptability;

[0118] The initial trajectory planning module is used to construct the main variable evolution tensor and risk nested evolution diagram based on a unified monitoring data set. It combines the main variable trigger frequency with the path cost function construction strategy to generate the initial trajectory planning sequence for the UAV formation and achieve the minimum cost path allocation under multi-objective mission conditions;

[0119] The trajectory execution and data acquisition module is used to control the UAV formation to conduct inspection flights in a timely and orderly manner according to the initial trajectory planning sequence. During the flight, the UAV flight trajectory data, attitude information and environmental status data are collected in real time, and a dual-channel flight-environment data set is generated in a spatiotemporal binding manner.

[0120] The fault-tolerant control module is used to compare the actual flight trajectory with the initial trajectory planning sequence, the current environmental state and the unified monitoring data set, and construct a consistency error function. When the error exceeds the threshold, the control mechanism based on the fractional-order sliding mode fault-tolerant switching and progressive degradation strategy is triggered, and the fault-tolerant control command is output;

[0121] The main variable closed-loop optimization module is used to call the main variable evolution closed-loop mechanism according to the fault-tolerant control instructions and the unified monitoring data set, dynamically update the main variable evolution tensor and path cost function, re-evaluate the path risk and evolution trend of each node, and output the optimized formation trajectory planning scheme and the corresponding structured warning decision results after meeting the convergence conditions;

[0122] The early warning decision output module is used to receive the early warning decision results generated by the main variable closed-loop optimization module, and output structured early warning information including high, medium and low risk level labels, corresponding node numbers and track task adjustment suggestions, and establish associations with historical task execution records for subsequent task maintenance and scheduling optimization.

[0123] In this embodiment, the modules are connected through the following methods:

[0124] S1. Synchronously collect surface settlement time series and crack evolution characteristic data of the goaf through the ground sensor node network to form the initial observation data set;

[0125] S2, input the initial observation data set into the fractal small-world network coding consensus algorithm for fusion processing, and output a unified monitoring data set;

[0126] S3. Based on the unified monitoring data set, the main variable evolution closed-loop mechanism is used to generate the initial trajectory planning sequence of the UAV formation;

[0127] S4. Control the UAV formation to execute the inspection route according to the initial trajectory planning sequence, and collect formation flight trajectory data and environmental status data in real time;

[0128] S5. Based on the real-time collected data, calculate the consistency error between the flight trajectory and the initial trajectory planning sequence, and the environmental state and the unified monitoring data set; when the consistency error exceeds the preset threshold, generate a warning signal and trigger the fractional-order sliding mode fault-tolerant switching and progressive degradation strategy, and output fault-tolerant control instructions;

[0129] S6. Based on the fault-tolerant control instructions and the unified monitoring data set, the main variable evolution closed-loop mechanism is called to update the trajectory cost function and the main variable evolution tensor state, and the path risk and evolution trend of each node are dynamically re-evaluated. After the convergence conditions are met, the optimized formation trajectory planning scheme and the corresponding warning decision results are output.

[0130] In this embodiment, S1 includes the following specific steps:

[0131] S11. Deploy multiple ground sensor nodes, distributed in the active settlement areas and crack concentration areas of the goaf. Each sensor node is equipped with a settlement measurement unit and a crack monitoring unit.

[0132] S12. During each monitoring period, record the surface vertical displacement data and crack opening width data collected by each node, and attach the corresponding time tag and node number;

[0133] S13, the settlement amount and crack parameters collected by each node in the current period are combined into a two-dimensional observation vector, which includes spatial identification and time identification information;

[0134] S14. Summarize all observation vectors in the current period by node number to form a two-dimensional original observation matrix, where rows represent different nodes and columns represent settlement values ​​and crack values ​​of corresponding nodes;

[0135] S15. Perform numerical normalization on the original observation matrix, unify the dimensions, and synchronously calibrate the time tags to generate an initial observation data set.

[0136] In this embodiment, the fractal small-world network coding consensus algorithm includes the following specific steps:

[0137] Constructing a task-sensitive fractal small-world network topology in the goaf monitoring area includes the following sub-steps:

[0138] Obtain a node set, classify the nodes according to their functional types to form a functional hierarchical structure, classify the surface subsidence monitoring nodes into the inner nested substructure, and classify the drone image recognition nodes into the outer nested substructure;

[0139] Establish a circular local connection relationship based on spatial proximity within each fractal level to generate an adjacent edge set ε local ;

[0140] According to the historical geological risk level r of the area where each node is locatedj , select remote nodes from different fractal levels to build remote connection edge sets ε remote , where node v i With node v j The probability of long-range connection between ij ;

[0141] At the initialization time t0, for each node v i Construct the observation state vector: where d i (t0) is the node v i The vertical settlement value of the ground collected at time t0, c i (t0) is the crack opening width, s i (t0) is the node structure attribute label, including function type, deployment level and task priority weight;

[0142] Calculate the state covariance matrix Σ of each node i , represents the data distribution structure of the current node in the current observation period, and defines the structural consistency weight between it and the adjacent nodes as:

[0143]

[0144] in,‖·‖ F represents the Frobenius norm, is the set of adjacent nodes of the node;

[0145] Set the main variable dimension to the surface settlement value component d in the node observation state vector i (t k ), calculate its rate of change between two consecutive sampling periods: δ i (t k )=|d i (t k )-d i (t k-1 )|, and set the dynamic threshold function:

[0146] θ(t k )=θ0·(1-β·R(t k ));

[0147] Among them, θ0 is the initial trigger threshold, β is the task risk adjustment coefficient, R(t k ) is the risk level of the system’s current task;

[0148] If the node v i Satisfy δ i (t k )>θ(t k), the node state update is triggered, and the new state vector is calculated according to the following network coding fusion formula:

[0149]

[0150] in, Represents the state value of the neighbor node in the kth iteration;

[0151] After each round of status update, the consensus termination conditions are determined. If any node meets one of the following conditions, the consensus iteration is terminated:

[0152] The state change amplitude is lower than the set convergence threshold ε, that is,

[0153] The mission risk level reaches or exceeds the warning threshold, that is, R(t k )≥R max ;

[0154] The number of iterations exceeds the maximum limit, that is, k ≥ K max ;

[0155] After consensus is completed, each node outputs the final fusion state Build a unified monitoring dataset:

[0156] In this embodiment, S3 includes the following specific steps:

[0157] S31, construct the main variable evolution tensor set, and number each monitoring node as v i , extract the main variable observation sequence within T consecutive time periods, including the sedimentation rate sequence, the crack expansion rate sequence and the image structure disturbance degree sequence, and reconstruct it into a three-dimensional tensor: in For node v i The main variable tensor is q, the number of main variable types is q, T is the time window length, and R is the spatial projection scale of the node adjacent area.

[0158] S32. Construct a risk nested evolution diagram based on the main variable tensor of each node in the unified monitoring data set The node set Corresponding to the set of monitoring nodes, the edge weight in the graph is defined as

[0159]

[0160] in, Represents node v i The observation value of the lth dimension of the main variable, K is the sliding window length, w ij (tk ) is the co-evolution edge weight of the main variable;

[0161] S33, using the norm of the state vector of each node in the evolution graph, set the trigger frequency of the local main variable and calculate the node v i The main variable evolution trigger period is:

[0162]

[0163] Where Δt i Represents node v i The local update period, Δt min is the minimum update period, γ is the control adjustment factor, h i (t k ) is the node state vector;

[0164] S34. Extract the node set that triggers the update, generate the track cost distribution map R(x,y) based on the risk projection of the main variable tensor in two-dimensional space, and construct the path cost function of each drone as

[0165]

[0166] in, represents the point sequence of the UAV numbered j in the initial path planning, λ is the risk adjustment coefficient, Cost j is the corresponding path cost;

[0167] S35. Optimize the allocation of paths based on the minimum cost principle, and record the minimum cost path set as

[0168] The path set is the initial trajectory planning sequence of the UAV formation.

[0169] In this embodiment, S4 includes the following specific steps:

[0170] S41. Based on the initial trajectory planning sequence, the trajectory point sets of each UAV are issued in sequence to determine the three-dimensional spatial route that each UAV needs to execute, clarify the position parameters of the starting point, target point, and intermediate control points, and set the corresponding flight scheduling order and time interval;

[0171] S42. Control each UAV to enter the target flight segment according to the planned sequence. Based on the time-triggered control strategy, maintain the speed and attitude consistent with the track point set at continuous time, perform spatial position updates at fixed intervals, and record the current flight position, velocity vector, and attitude angle parameters in real time.

[0172] S43. Synchronously collect environmental status data corresponding to the current flight position at each moment, including air temperature, humidity, wind speed, particle concentration, electromagnetic disturbance, remaining battery ratio, and image light intensity information. The environmental status data is structured and stored in a manner that binds time tags and spatial location information;

[0173] S44, performing time synchronization fusion on the flight status data and environmental status data of each UAV during the execution of the current track segment, generating a flight-environment dual-channel time series dataset of the continuous inspection segment, and marking the current mission segment number and track segment type label;

[0174] S45. The flight trajectory data and environmental status data collected by all UAVs in the formation during the current inspection cycle are aggregated to generate a complete inspection execution data set.

[0175] In this embodiment, S5 includes the following specific steps:

[0176] S51. Based on the inspection execution data set, obtain the current time t of the flight mission k The actual flight trajectory collection The initial trajectory planning sequence of the generated UAV formation And construct the trajectory consistency error function based on the Euclidean distance:

[0177]

[0178] Where N is the number of drones, p a,i (t k ) and p r,i (t k ) represent the actual and reference positions of the i-th UAV respectively;

[0179] S52, synchronously collect task environment state data vector E(t k ), extract the reference environment state vector E at the same moment from the unified monitoring data set * (t k ), construct the state consistency error function: ∈ e (t k )=‖E(t k )-E * (t k )‖2;

[0180] S53, the trajectory error ∈ p (t k ) and state error ∈ e (t k ) is fused into the total consistency error:

[0181] ∈(t k )=γ1·∈p (tk)+γ2·∈ e (t k );

[0182] Wherein, γ1 and γ2 are preset weighting coefficients, satisfying γ1+γ2=1;

[0183] S54, when ∈(t k )>∈ th When the warning signal is output, the fault-tolerant switching and progressive degradation strategy of the fractional-order sliding mode is triggered, and the task node set is The main sliding mode controller performs degradation switching and control mode switching, and dynamically adjusts the control input vector u(t k );

[0184] S55, output adjusted fault-tolerant control instruction set And update the execution status in the controller lineage structure according to the current degradation path node.

[0185] In this embodiment, the fractional-order sliding mode fault-tolerant switching and progressive degradation strategy includes the following specific steps:

[0186] Build a collection of task nodes At each node Collect the main variable evolution tensor And combined with the risk nested evolution diagram to generate the local risk intensity value ρ i (t), define the risk perception function based on task disturbance and environmental uncertainty:

[0187]

[0188] Among them, δ0 is the initial scale of the sliding mode domain, β is the task risk adjustment coefficient, Represents the gradient norm of the tensor in the time axis direction;

[0189] At each node v i Construct a fractional sliding surface:

[0190]

[0191] in, represents the Caputo fractional derivative, order α i (t) Dynamically adjust with risk perception function; e i (t) is the difference between the target track and the actual track; λ i (t) is the environmental state sensitive adjustment gain;

[0192] Aggregate the sliding mode control surfaces of each node to form a regional control law set The sliding mode approach domain is defined as: And according to the ranking results of risk perception functions in each region, a weighted combination of regional control laws is performed:

[0193] Constructing a degenerate lineage tree Each node M j represents a control model with reduced precision, and the degradation path is determined by the following objective function:

[0194]

[0195] Among them, C j (t) is the task cost, F j (t) is the controller's remaining execution capability indicator, R j (t) is the current environment robustness score, η1, η2, η3 are normalized weights;

[0196] At each moment, according to the currently selected degradation model M * (t) Switch the control law set and replace the original sliding mode surface with the weak sliding mode structure in the corresponding degenerate node to perform continuous degradation control;

[0197] Perform a backtracking update operation on the degradation sequence. When the task error is lower than the tolerance range and the risk function continuously decreases for more than the set period, it automatically performs an upward transfer back to the previous layer node M j-1 , and reconstruct the sliding mode controller parameters.

[0198] In this embodiment, S6 includes the following specific steps:

[0199] S61, set the fault-tolerant control instruction set With the unified monitoring dataset as the joint input, for each trigger node in the unified monitoring dataset, the main variable evolution closed-loop mechanism is called to update the main variable tensor and calculate the risk intensity. The risk intensity is defined as the product of the norm increment of the node main variable tensor in the last two cycles and the current path error weight.

[0200] S62、The risk intensity R i The node path cost C is obtained by weighted summing the node environmental disturbance index, power threshold and historical track deviation within the mission window. i ;

[0201] S63. For each UAV, the node path costs of all path points are weighted and accumulated to obtain the total cost of the candidate track. The candidate path point sequence is then subjected to the following steps: adjusting the path point order to minimize the total cost; inserting risk avoidance points into high-risk sections; and deleting invalid redundant sections.

[0202] S64, detect the change amplitude of the main variable tensor in multiple consecutive scheduling cycles Total path cost ΔC = |C (k) -C (k-1) | and node risk sequence The iteration is terminated when any of the following conditions is met: ΔX<ε x , ΔC<ε c , ΔR<ε r , reaching the maximum number of iterations K max , where ε x is the convergence threshold of the main variable tensor, ε c is the convergence threshold of the total path cost, ε r is the convergence threshold of the node risk sequence;

[0203] S65, output the optimized formation trajectory planning scheme at the current moment; at the same time, based on the risk intensity R of each node i The evolution trend of the main variables at the time of convergence termination generates a structured warning decision result, including: node risk level label, defined as: high R i ≥T h , Middle T l ≤R i <T h , low R i <T l , Warning node number list, formation task adjustment suggestions, where T h ,T l is the risk level threshold.

[0204] Example 1:

[0205] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the automated inspection and risk warning measurement tasks in the key observation area of ​​subsidence disasters in a certain goaf area. The test area is located above the old mining layer. The historical goaf layer is more than 300 meters deep. There are active cracks and slope deformation traces on the surface. The test period is from March 1, 2025 to March 30, 2025. 72 ground main variable sensor nodes are deployed in the test area, covering settlement monitoring, crack width measurement and surface temperature and humidity disturbance collection. The system uses three modified six-rotor drones with a flight time of 50 minutes, equipped with high-precision GPS and inertial navigation modules, and performs three rounds of full coverage tasks every day to complete dynamic inspections and risk feedback in different areas.

[0206] The traditional approach employed a fixed-path formation and single-threshold early warning strategy during the initial testing phase. In the early morning hours of the seventh day of testing, the subsidence rate at nodes M26, M27, and M31 suddenly increased to over 5.2 mm / day. However, the traditional system's trajectory did not cover this area, resulting in a seven-hour response delay and a missed false alarm. The crack opening had expanded to 4.7 cm by the time the drone made its first go-around, and the primary variable sequence fluctuated significantly, necessitating a subsequent trajectory reset.

[0207] To address these issues, the research team integrated their proposed multi-source monitoring and closed-loop control system, based on fractal small-world network consensus, into the task chain. The system receives sampling sequences from each node in real time and iteratively reconstructs a unified monitoring dataset through network coding. Simultaneously, it constructs a principal variable evolution tensor and a risk nested evolution graph to drive the initial trajectory generation module for path planning. During flight, dual channels collect flight trajectory and environmental disturbance data, calculating trajectory-state consistency errors in real time. Once the error function exceeds a set threshold, a fractional-order sliding mode-based fault-tolerant switching mechanism is immediately triggered, enabling the degradation control model and initiating a risk avoidance process.

[0208] During system operation, the system recorded nine sudden major variable events, four of which were high-level risks. For each event, path reconstruction was completed within 10 minutes, with 13 new avoidance waypoints added to ensure uninterrupted mission continuity. In particular, during the night of the 19th day of testing, when wind disturbances were severe, the system, through its node collaborative prediction mechanism, determined that a risk concentration existed in the northeastern M33M35 region. It proactively corrected its trajectory away from the fissure zone, completing the avoidance maneuver approximately 28 minutes before the risk point erupted, successfully avoiding the path from crossing the high-disturbance area.

[0209] The system performed a total of 270 drone missions, integrated and processed approximately 142,000 sets of primary variable data, and output 38 structured warning messages. The system achieved a 94.3% accuracy rate for real high-risk event warnings, with an average trajectory reconstruction response delay of only 3.6 seconds. After comparing the system with traditional path control systems, the research team analyzed typical indicators, with the following results shown in the table below:

[0210] In typical scenarios involving sudden subsidence, crack expansion, and the coupling of disturbances, the present invention demonstrates significant advantages in primary variable identification accuracy, path reconstruction speed, risk avoidance success rate, and false alarm rate. In particular, in situations with wind speeds exceeding 6 m / s or sudden changes at multiple nodes, traditional systems often experience path crossings of risk points or delayed warnings due to response lags. However, the present invention achieves highly robust task scheduling through a closed-loop mechanism combining evolutionary tensors and fault-tolerant control.

[0211] The following table compares the core performance dimensions of the system of the present invention and traditional methods under different risk types during the 30-day test:

[0212] Table 1 Performance comparison between the present invention and traditional methods under different risk scenarios

[0213]

[0214]

[0215] As shown in the comparative results in Table 1, in the "sudden change in settlement" scenario, the present invention improved the accuracy of primary variable identification from 67.5% to 91.2%, the risk avoidance success rate from 63.2% to 88.5%, and the track response time was reduced to less than one-third of the original. Under wind disturbance coupling, the false alarm rate of the traditional system was as high as 18.2%, while that of the present invention was only 6.0%, significantly improving stability.

[0216] In summary, the present invention demonstrates high efficiency in structural continuity warning, dynamic task adjustment, and anomaly avoidance. It is suitable for unmanned geological disaster risk monitoring scenarios that require high frequency, high timeliness, and low false alarm rate, and has good engineering deployment adaptability and practical value.

[0217] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A goaf area drone inspection system based on monitoring and early warning, characterized by: include: The observation data acquisition module is used to collect time series data on surface subsidence and crack evolution characteristics in the goaf. By deploying ground sensor nodes in active subsidence areas and areas with concentrated cracks, it obtains observation data on primary variables such as vertical displacement and crack opening width within each monitoring cycle, and generates an initial observation data set with time tags and node numbers. The data fusion module is used to input the initial observation data set into the fractal small-world network coding consensus algorithm for fusion processing, construct a task-sensitive fractal topology structure, perform iterative consensus calculation on the node state vector, and output a unified monitoring data set that meets structural consistency and risk adaptability; The initial trajectory planning module is used to construct the main variable evolution tensor and risk nested evolution diagram based on a unified monitoring data set. It combines the main variable trigger frequency with the path cost function construction strategy to generate the initial trajectory planning sequence for the UAV formation and achieve the minimum cost path allocation under multi-objective mission conditions; The trajectory execution and data acquisition module is used to control the UAV formation to conduct inspection flights in a timely and orderly manner according to the initial trajectory planning sequence. During the flight, the UAV flight trajectory data, attitude information and environmental status data are collected in real time, and a dual-channel flight-environment data set is generated in a spatiotemporal binding manner. The fault-tolerant control module is used to compare the actual flight trajectory with the initial trajectory planning sequence, the current environmental state and the unified monitoring data set, and construct a consistency error function. When the error exceeds the threshold, the control mechanism based on the fractional-order sliding mode fault-tolerant switching and progressive degradation strategy is triggered, and the fault-tolerant control command is output; The main variable closed-loop optimization module is used to call the main variable evolution closed-loop mechanism according to the fault-tolerant control instructions and the unified monitoring data set, dynamically update the main variable evolution tensor and path cost function, re-evaluate the path risk and evolution trend of each node, and output the optimized formation trajectory planning scheme and the corresponding structured warning decision results after meeting the convergence conditions; The early warning decision output module is used to receive the early warning decision results generated by the main variable closed-loop optimization module, and output structured early warning information including high, medium and low risk level labels, corresponding node numbers and track task adjustment suggestions, and establish associations with historical task execution records for subsequent task maintenance and scheduling optimization.

2. The goaf area drone inspection system based on monitoring and early warning according to claim 1 is characterized in that: The modules are implemented as follows: S1. Synchronously collect surface settlement time series and crack evolution characteristic data of the goaf through the ground sensor node network to form the initial observation data set; S2, input the initial observation data set into the fractal small-world network coding consensus algorithm for fusion processing, and output a unified monitoring data set; S3. Based on the unified monitoring data set, the main variable evolution closed-loop mechanism is used to generate the initial trajectory planning sequence of the UAV formation; S4. Control the UAV formation to execute the inspection route according to the initial trajectory planning sequence, and collect formation flight trajectory data and environmental status data in real time; S5. Based on the real-time collected data, calculate the consistency error between the flight trajectory and the initial trajectory planning sequence and the environmental state and the unified monitoring data set; When the consistency error exceeds the preset threshold, a warning signal is generated and the fractional-order sliding mode fault-tolerant switching and progressive degradation strategy are triggered to output fault-tolerant control instructions; S6. Based on the fault-tolerant control instructions and the unified monitoring data set, the main variable evolution closed-loop mechanism is called to update the trajectory cost function and the main variable evolution tensor state, and the path risk and evolution trend of each node are dynamically re-evaluated. After the convergence conditions are met, the optimized formation trajectory planning scheme and the corresponding warning decision results are output.

3. The goaf area drone inspection system based on monitoring and early warning according to claim 2 is characterized in that: The S1 includes the following specific steps: S11. Deploy multiple ground sensor nodes distributed in the active settlement areas and crack concentration areas of the goaf. Each sensor node is equipped with a settlement measurement unit and a crack monitoring unit. S12. During each monitoring period, record the surface vertical displacement data and crack opening width data collected by each node, and attach the corresponding time tag and node number; S13, the settlement amount and crack parameters collected by each node in the current period are combined into a two-dimensional observation vector, which includes spatial identification and time identification information; S14. Summarize all observation vectors in the current period by node number to form a two-dimensional original observation matrix, where rows represent different nodes and columns represent settlement values ​​and crack values ​​of corresponding nodes; S15. Perform numerical normalization on the original observation matrix, unify the dimensions, and synchronously calibrate the time tags to generate an initial observation data set.

4. The fractal small-world network coding consensus algorithm according to claim 2, characterized in that: The specific steps include: Constructing a task-sensitive fractal small-world network topology in the goaf monitoring area includes the following sub-steps: Obtain a node set, classify the nodes according to their functional types to form a functional hierarchical structure, classify the surface subsidence monitoring nodes into the inner nested substructure, and classify the drone image recognition nodes into the outer nested substructure; Establish a circular local connection relationship based on spatial proximity within each fractal level to generate an adjacent edge set ε local ; According to the historical geological risk level r of the area where each node is located j , select remote nodes from different fractal levels to build remote connection edge sets ε remote , where node v i With node v j The probability of long-range connection between ij ; At the initialization time t0, for each node v i Construct the observation state vector: where d i (t0) is the node v i The vertical settlement value of the ground collected at time t0, c i (t0) is the crack opening width, s i (t0) is the node structure attribute label, including function type, deployment level and task priority weight; Calculate the state covariance matrix Σ of each node i , represents the data distribution structure of the current node in the current observation period, and defines the structural consistency weight between it and the adjacent nodes as: in,‖·‖ F represents the Frobenius norm, is the set of adjacent nodes of the node; Set the main variable dimension to the surface settlement value component d in the node observation state vector i (t k ), calculate its rate of change between two consecutive sampling periods: δ i (t k )=|d i (t k )-d i (t k-1 )|, and set the dynamic threshold function: θ(t k )=θ0·(1-β·R(t k )); Among them, θ0 is the initial trigger threshold, β is the task risk adjustment coefficient, R(t k ) is the risk level of the system’s current task; If the node v i Satisfy δ i (t k )>θ(t k ), the node state update is triggered, and the new state vector is calculated according to the following network coding fusion formula: in, Represents the state value of the neighbor node in the kth iteration; After each round of status update, the consensus termination conditions are determined. If any node meets one of the following conditions, the consensus iteration is terminated: The state change amplitude is lower than the set convergence threshold ε, that is, The mission risk level reaches or exceeds the warning threshold, that is, R(t k )≥R max ; The number of iterations exceeds the maximum limit, that is, k ≥ K max ; After consensus is completed, each node outputs the final fusion state Build a unified monitoring dataset:

5. The goaf area drone inspection system based on monitoring and early warning according to claim 2 is characterized in that: The S3 includes the following specific steps: S31, construct the main variable evolution tensor set, and number each monitoring node as v i , extract the main variable observation sequence within T consecutive time periods, including the sedimentation rate sequence, the crack expansion rate sequence and the image structure disturbance degree sequence, and reconstruct it into a three-dimensional tensor: in For node v i The main variable tensor is q, the number of main variable types is q, T is the time window length, and R is the spatial projection scale of the node adjacent area. S32. Construct a risk nested evolution diagram based on the main variable tensor of each node in the unified monitoring data set The node set Corresponding to the set of monitoring nodes, the edge weight in the graph is defined as in, Represents node v i The observation value of the lth dimension of the main variable, K is the sliding window length, w ij (t k ) is the co-evolution edge weight of the main variable; S33, using the norm of the state vector of each node in the evolution graph, set the trigger frequency of the local main variable and calculate the node v i The main variable evolution trigger period is: Where Δt i Represents node v i The local update period, Δt min is the minimum update period, γ is the control adjustment factor, h i (t k ) is the node state vector; S34. Extract the node set that triggers the update, generate the track cost distribution map R(x,y) based on the risk projection of the main variable tensor in two-dimensional space, and construct the path cost function of each drone as in, represents the point sequence of the UAV numbered j in the initial path planning, λ is the risk adjustment coefficient, Cost j is the corresponding path cost; S35. Optimize the allocation of paths based on the minimum cost principle, and record the minimum cost path set as The path set is the initial trajectory planning sequence of the UAV formation.

6. The goaf area drone inspection system based on monitoring and early warning according to claim 2 is characterized in that: The S4 includes the following specific steps: S41. Based on the initial trajectory planning sequence, the trajectory point sets of each UAV are issued in sequence to determine the three-dimensional spatial route that each UAV needs to execute, clarify the position parameters of the starting point, target point, and intermediate control points, and set the corresponding flight scheduling order and time interval; S42. Control each UAV to enter the target flight segment according to the planned sequence. Based on the time-triggered control strategy, maintain the speed and attitude consistent with the track point set at continuous time, perform spatial position updates at fixed intervals, and record the current flight position, velocity vector, and attitude angle parameters in real time. S43. Synchronously collect environmental status data corresponding to the current flight position at each moment, including air temperature, humidity, wind speed, particle concentration, electromagnetic disturbance, remaining battery ratio, and image light intensity information. The environmental status data is structured and stored in a manner that binds time tags and spatial location information; S44, performing time synchronization fusion on the flight status data and environmental status data of each UAV during the execution of the current track segment, generating a flight-environment dual-channel time series dataset of the continuous inspection segment, and marking the current mission segment number and track segment type label; S45. The flight trajectory data and environmental status data collected by all UAVs in the formation during the current inspection cycle are aggregated to generate a complete inspection execution data set.

7. The goaf inspection system based on monitoring and early warning by drones according to claim 2 is characterized in that: The S5 includes the following specific steps: S51. Based on the inspection execution data set, obtain the current time t of the flight mission k The actual flight trajectory collection The initial trajectory planning sequence of the generated UAV formation And construct the trajectory consistency error function based on the Euclidean distance: Where N is the number of drones, p a,i (t k ) and p r,i (t k ) represent the actual and reference positions of the i-th UAV respectively; S52, synchronously collect task environment state data vector E(t k ), extract the reference environment state vector E at the same moment from the unified monitoring data set * (t k ), construct the state consistency error function: ∈ e (t k )=‖E(t k )-E * (t k )‖2; S53, the trajectory error ∈ p (t k ) and state error ∈ e (t k ) is fused into the total consistency error: ∈(t k )=γ1·∈ p (t k )+γ2·∈ e (t k ); Wherein, γ1 and γ2 are preset weighting coefficients, satisfying γ1+γ2=1; S54, when ∈(t k )>∈ th When the warning signal is output, the fault-tolerant switching and progressive degradation strategy of the fractional-order sliding mode is triggered, and the task node set is The main sliding mode controller performs degradation switching and control mode switching, and dynamically adjusts the control input vector u(t k ); S55, output adjusted fault-tolerant control instruction set And update the execution status in the controller lineage structure according to the current degradation path node.

8. The fractional-order sliding mode fault-tolerant switching and progressive degradation strategy according to claim 2, characterized in that: The specific steps include: Build a collection of task nodes At each node Collect the main variable evolution tensor And combined with the risk nested evolution diagram to generate the local risk intensity value ρ i (t), define the risk perception function based on task disturbance and environmental uncertainty: Among them, δ0 is the initial scale of the sliding mode domain, β is the task risk adjustment coefficient, Represents the gradient norm of the tensor in the time axis direction; At each node v i Construct a fractional sliding surface: in, represents the Caputo fractional derivative, order α i (t) Dynamically adjust with risk perception function; e i (t) is the difference between the target track and the actual track; λ i (t) is the environmental state sensitive adjustment gain; Aggregate the sliding mode control surfaces of each node to form a regional control law set The sliding mode approach domain is defined as: And according to the ranking results of risk perception functions in each region, a weighted combination of regional control laws is performed: Constructing a degenerate lineage tree Each node M j represents a control model with reduced precision, and the degradation path is determined by the following objective function: Among them, C j (t) is the task cost, F j (t) is the controller's remaining execution capability indicator, R j (t) is the current environment robustness score, η1, η2, η3 are normalized weights; At each moment, according to the currently selected degradation model M * (t) Switch the control law set and replace the original sliding mode surface with the weak sliding mode structure in the corresponding degenerate node to perform continuous degradation control; Perform a backtracking update operation on the degradation sequence. When the task error is lower than the tolerance range and the risk function continuously decreases for more than the set period, it automatically performs an upward transfer back to the previous layer node M j-1 , and reconstruct the sliding mode controller parameters.

9. The goaf area drone inspection system based on monitoring and early warning according to claim 2 is characterized in that: The S6 includes the following specific steps: S61, set the fault-tolerant control instruction set With the unified monitoring dataset as the joint input, for each trigger node in the unified monitoring dataset, the main variable evolution closed-loop mechanism is called to update the main variable tensor and calculate the risk intensity. The risk intensity is defined as the product of the norm increment of the node main variable tensor in the last two cycles and the current path error weight. S62、The risk intensity R i The node path cost C is obtained by weighted summing the node environmental disturbance index, power threshold and historical track deviation within the mission window. i ; S63. For each UAV, weighted sum the node path costs of all path points to obtain the total cost of the candidate track. For the candidate path point sequence, perform the following operations: adjust the path point order to minimize the total cost; insert risk avoidance points into high-risk sections; Delete invalid redundant segments; S64, detect the change amplitude of the main variable tensor in multiple consecutive scheduling cycles Total path cost ΔC = |C (k) -C (k-1) | and node risk sequence The iteration is terminated when any of the following conditions is met: ΔX<ε x , ΔC<ε c , ΔR<ε r , reaching the maximum number of iterations K max , where ε x is the convergence threshold of the main variable tensor, ε c is the convergence threshold of the total path cost, ε r is the convergence threshold of the node risk sequence; S65, output the optimized formation trajectory planning scheme at the current moment; at the same time, based on the risk intensity R of each node i The evolution trend of the main variables at the time of convergence termination generates a structured warning decision result, including: node risk level label, defined as: high R i ≥T h , Middle T l ≤R i <T h , low R i <T l , Warning node number list, formation task adjustment suggestions, where T h ,T l is the risk level threshold.

Images