Power plant pollution plume tracking method based on cooperation of unmanned aerial vehicle remote sensing and ground monitoring

Abstract

The application discloses a power plant pollution plume tracking method based on cooperation of unmanned aerial vehicle remote sensing and ground monitoring. The method reconstructs a phase space of a concentration time sequence collected and calculates a correlation dimension, drives a dynamic variable-pitch multi-dimensional sensing array to adjust a physical baseline length according to the correlation dimension, performs simplex homology group algebra calculation on a concentration point set, extracts zero-order and first-order Betti numbers as tracking judgment basis, triggers a zero-concentration fault identification mechanism for clean air, establishes a spatial topological principal direction vector based on a spatial covariance matrix of a main connected branch set and guides straight and level flight, verifies cross-fault tracking through homology mapping after regaining an effective signal, and finally outputs a spatial topological diffusion track of the pollution plume. The application realizes physical linkage of sampling resolution and flow field dynamic characteristics, overcomes the target loss problem caused by guidance failure under complex weather conditions, guarantees the continuity of the tracking path on the spatial topological level, and achieves the purpose of accurate tracking of complex pollution plumes.

Description

Technical Field

[0001] This invention relates to the field of atmospheric environment monitoring and UAV application technology, specifically to a method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring. Background Technology

[0002] The exhaust gases emitted by thermal power plants form continuous pollution plumes during their diffusion process, directly impacting the atmospheric environmental quality of the surrounding and downwind areas. In order to accurately assess the degree of environmental pollution, understand the transmission patterns of pollutants, and accurately trace their sources, it is necessary to track the three-dimensional spatial distribution characteristics and diffusion trajectory of pollution plumes in real time. In recent years, drones, with their high mobility, flexible operating altitude, and wide coverage, have become an important technical means for industrial emission monitoring when equipped with various gas sensing devices for high-altitude plume detection.

[0003] In existing conventional UAV plume tracking operations, the common approach is to install multiple gas sensor arrays on the fuselage of multi-rotor or fixed-wing UAVs. During the operation, the UAV cruises and samples in the target airspace according to a pre-planned gridded route, or directly controls the aircraft to move from a low concentration area to a high concentration area based on the single concentration gradient data collected in real time by the sensors. In this operation mode, ground monitoring stations mostly serve as auxiliary facilities to provide reference for regular meteorological conditions or background concentrations at fixed locations, and aerial detection relies entirely on the fixed spatial acquisition structure of airborne sensors to obtain plume concentration information.

[0004] However, pollution plumes in the atmosphere are affected by wind speed, wind shear, and turbulence in complex terrain. Their internal flow field dynamics and spatial diffusion structure exhibit highly nonlinear and chaotic characteristics. Existing UAV-borne multi-sensor array structures are fixed, and the physical spatial baseline distance between each detection probe is not adjustable at the hardware level. This results in the spatial sampling resolution of the entire system being in a static locked state. This static hardware sampling structure is completely unable to adapt to the dynamic changes in the complexity of the local flow field. When the nonlinear diffusion of the pollution plume causes violent fluctuations in its internal structure, the fixed baseline sensor cannot adaptively match the evolution of the flow field, and thus cannot effectively obtain the true and accurate spatial distribution characteristics of the pollution plume. Summary of the Invention

[0005] The technical problem to be solved by this invention is that the physical spatial baseline distance of the UAV-borne multi-sensor array is fixed and cannot be adjusted in the existing technology, which causes the spatial sampling resolution of the system to be in a static locked state, making it impossible to adapt to the dynamic changes in the local flow field dynamics and complexity of the pollution plume, and thus making it difficult to effectively obtain the true spatial distribution characteristics of the pollution plume. This invention provides a pollution plume tracking method for thermal power plants based on the collaboration of UAV remote sensing and ground monitoring.

[0006] This invention provides a method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring, comprising the following steps:

[0007] Step S110: The environmental parameter acquisition module of the ground monitoring station acquires the time series of characteristic pollutant concentrations within a preset time window and sends it to the edge computing node of the same station.

[0008] In step S120, the edge computing node reconstructs the time series of characteristic pollutant concentrations into a high-dimensional phase space state vector, extracts the correlation dimension of the high-dimensional phase space state vector distribution, and sends it to the airborne computing unit of the airborne detection subsystem through a wireless communication link.

[0009] In step S130, the airborne computing unit substitutes the received correlation dimension into the control equation to convert it into a mechanical displacement command, which is then sent to the servo motor driver of the dynamic variable pitch multidimensional sensor array. The servo motor driver drives the mechanical telescopic link to perform a retracting or extending action according to the mechanical displacement command.

[0010] Step S140: Under the current mechanical link length configuration, the dynamic variable-distance multi-dimensional sensing array collects a set of discrete points of three-dimensional spatial concentration in the airspace where the UAV is located, and transmits them back to the airborne computing unit via the internal bus.

[0011] In step S150, the airborne computing unit performs simple homology group algebra calculation on the discrete point set of three-dimensional spatial concentration, extracts the zero-order Betti number and the first-order Betti number, and uses the extracted zero-order Betti number and the first-order Betti number as the basis for determining whether the multi-rotor UAV platform has entered the clean air fault region and for performing cross-fault trajectory planning. The unit then sends flight guidance commands to the flight control board to control the flight of the multi-rotor UAV platform, so as to dynamically adjust the sampling position and continuously update the effective spatial sampling point set, realize continuous cross-fault tracking of the pollution plume from the thermal power plant, and output the spatial topological diffusion trajectory of the pollution plume.

[0012] The optimal time delay parameter is determined using the mutual information method, and the minimum embedding dimension parameter is determined using the spurious nearest neighbor method. Edge computing nodes calculate the correlation integral and construct a double logarithmic coordinate mapping relationship. The correlation dimension is determined by performing linear fitting in the scale-free region. The correlation dimension is used to evaluate the nonlinear structural complexity of the pollution plume flow field. The airborne computing unit adjusts the extension length of the mechanically retractable link according to the mapping relationship between the correlation dimension and the critical threshold, so that the spatial sampling resolution of the sensor adapts to the evolution state of the local flow field.

[0013] The airborne computing unit uses the Lyapunov exponent to solve the dynamic filtering threshold and constructs a simple complex for the effective spatial sampling point set. It solves the matrix rank by performing elementary transformations on the boundary matrices of each dimension and calculates the Betti number based on the rank-to-zero theorem. The zero-order Betti number and the first-order Betti number respectively characterize the connectivity and porosity of the polluted area. The above topological invariants provide an algebraic basis for determining the continuity of the polluted area.

[0014] To address the clean air gap in the polluted plume, the above method employs a zero-concentration gap identification mechanism. When the digital concentration signal from the detection probe remains below the background noise baseline, it is determined that the multi-rotor UAV platform has entered the clean air gap region. The system extracts the independent connected branch containing the largest number of effective spatial sampling points as the main connected branch set. The onboard computing unit constructs the third-order spatial covariance matrix of the main connected branch set and performs eigenvalue decomposition. By calculating the inner product of the initial eigenvector and the historical flight velocity vector, directional ambiguity is eliminated, and the main spatial topological direction vector is established. The flight control board guides the multi-rotor UAV platform to perform straight-line level flight along the main spatial topological direction.

[0015] After crossing the fault and re-acquiring a valid signal, the system calculates the vertical distance from the geometric center of the new connected branch space to the target space line. If the vertical distance is less than the matching threshold and the new zero-order Betti number representation structure is complete, the system confirms that the cross-fault tracking is successful and restores the dynamic tracking logic control closed loop. The above verification process eliminates interference from non-target sources and ensures the continuity of the tracking track at the spatial topology level.

[0016] The present invention, by adopting the above technical solution, can bring the following beneficial effects:

[0017] 1. This invention reconstructs the high-dimensional phase space state vector and extracts the correlation dimension of the local flow field through ground edge computing nodes. The received correlation dimension is substituted into the nonlinear control equation by the airborne computing unit and converted into mechanical displacement commands. This drives the mechanically retractable linkage of the dynamic variable pitch multidimensional sensor array to perform contraction or expansion actions to change the physical baseline distance between the detection probes. This realizes the dynamic adaptive adjustment of the system's sensor spatial sampling resolution and the nonlinear structural complexity of the pollution plume, ensuring effective acquisition of spatial distribution characteristics.

[0018] 2. This invention solves the dynamic filtering threshold on the three-dimensional spatial concentration discrete point set collected by a multi-rotor UAV platform and constructs a simple complex. It uses boundary operators to construct boundary matrices of each dimension and solves the matrix rank. Then, based on the rank-to-zero degree theorem, it extracts the zero-order Betti number and the first-order Betti number to represent the connectivity state of the polluted area. This enables the continuous guidance of the UAV path transition based on algebraic topological invariants when the concentration gradient information fails due to interference from complex meteorological conditions. This effectively overcomes the problem of track oscillation and target loss caused by relying solely on absolute concentration values ​​for determination.

[0019] 3. This invention triggers a zero-concentration tomography identification mechanism when the digital concentration signal is continuously lower than the background noise baseline. Based on the eigenvalue decomposition results of the spatial covariance matrix of the historical main connected branch set, it establishes the spatial topological main direction vector and guides forced straight-line level flight. When the signal is reacquired, the vertical distance from the spatial geometric center of the new connected branch to the target spatial straight line is calculated for coherence verification. This enables continuous tracking across the clean air zone along the topological inertia when the plume breaks up, avoiding interference from random noise or non-target sources. Finally, it accurately outputs the spatial topological diffusion trajectory of the polluted plume, and fully realizes the purpose of plume tracking in complex environments. Attached Figure Description

[0020] Figure 1 This is a system architecture diagram of the present invention;

[0021] Figure 2 This is a hardware structure diagram of the dynamic variable-distance multidimensional sensing array of the present invention;

[0022] Figure 3 This is a flowchart of the steps in the method for tracing pollution plumes in thermal power plants according to the present invention;

[0023] Figure 4 This is a flowchart of the control logic for zero-concentration fault identification and cross-fault coherence mapping tracking in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1

[0026] See attached document Figure 1-4 This embodiment provides a method for tracking pollution plumes from thermal power plants based on the collaboration of UAV remote sensing and ground monitoring, including the following steps:

[0027] Step S110: The environmental parameter acquisition module of the ground monitoring station continuously collects characteristic pollutant concentration data and sends it to the edge computing node at the same site;

[0028] In step S120, the edge computing node calculates the correlation dimension of the current local flow field based on the received time series data and sends it to the airborne computing unit of the airborne detection subsystem via a wireless communication link.

[0029] In step S130, the airborne computing unit substitutes the received correlation dimension into the control equation to convert it into a mechanical displacement command, which is then sent to the servo motor driver of the dynamic variable pitch multidimensional sensor array. The specific control equation is as follows:

[0030]

[0031] In the formula, Represents a dynamic variable pitch multidimensional sensing array in The length of the target physical baseline at any given time; This represents the maximum physical extended length allowed by the mechanical telescopic link; Represents the mechanical response sensitivity coefficient; Representing airborne computing power units in The number of associated dimensions received at any given time; Represents the critical correlation dimension threshold set by the system. The servo motor driver drives the mechanical telescopic link to perform contraction or expansion actions according to the mechanical displacement command.

[0032] Step S140: Under the current physical deployment configuration, the dynamic variable-distance multi-dimensional sensing array collects a set of three-dimensional density discrete points in the airspace where the UAV is located, and transmits them back to the onboard computing unit via the internal bus.

[0033] In step S150, the airborne computing unit performs simple homology group algebra calculation on the three-dimensional spatial concentration discrete point set, extracts the zero-order Betti number and the first-order Betti number, and uses the extracted zero-order Betti number and the first-order Betti number as the basis for determining whether the multi-rotor UAV platform has entered the clean air fault area and to perform cross-fault trajectory planning. The unit sends flight guidance commands to the flight control board to control the flight of the multi-rotor UAV platform, so as to dynamically adjust the sampling position and continuously update the effective spatial sampling point set, realize the continuous cross-fault tracking of the pollution plume of the thermal power plant, and output the spatial topological diffusion trajectory of the pollution plume.

[0034] Furthermore, the above method is implemented based on a coal-fired power plant pollution plume tracking system that combines UAV remote sensing and ground monitoring. This tracking system includes:

[0035] The system consists of a ground monitoring network subsystem and an aerial detection subsystem. The ground monitoring network subsystem is deployed in the area surrounding the thermal power plant and in the pre-defined downwind area, while the aerial detection subsystem is deployed on an unmanned aerial vehicle (UAV) platform and performs real-time detection tasks in the airspace in which it is located.

[0036] The ground monitoring network subsystem includes several fixed ground monitoring stations. Each ground monitoring station is equipped with an environmental parameter acquisition module and an edge computing node. The environmental parameter acquisition module includes a gas concentration sensor and a micro-meteorological sensor. The gas concentration sensor uses a non-dispersive infrared gas analyzer or a photoionization detector to collect time series data of the concentration of characteristic pollutants in thermal power plants. The micro-meteorological sensor uses an ultrasonic anemometer to collect three-dimensional wind speed and direction data at the measurement point location.

[0037] The edge computing node adopts an industrial-grade microcomputer based on ARM or x86 architecture, and is equipped with memory, solid-state storage and central processing unit. The edge computing node is connected to the environmental parameter acquisition module via RS485 bus or industrial Ethernet. The edge computing node receives concentration time series data and performs nonlinear dynamic calculations, and outputs correlation dimension.

[0038] The aerial detection subsystem includes a multi-rotor unmanned aerial vehicle platform, a dynamic variable-pitch multi-dimensional sensor array, an airborne computing unit, and a flight control board;

[0039] The dynamic variable pitch multidimensional sensor array is mounted on the bottom or arm of a multi-rotor UAV platform. The dynamic variable pitch multidimensional sensor array includes a central node, multiple outwardly extending mechanical telescopic links, and detection probes arranged at the ends of the links. The detection probes integrate a miniature gas sensor array. The mechanical telescopic links adopt an electronically controlled telescopic sleeve structure driven by a stepper motor or an electric push rod structure. The central node has a built-in servo motor driver. The servo motor driver receives external control signals and adjusts the extension length of each mechanical telescopic link, thereby changing the physical spatial baseline distance between each detection probe.

[0040] The airborne computing unit adopts an embedded computing module equipped with a graphics processor. The airborne computing unit establishes a data interaction connection with the dynamic variable pitch multidimensional sensor array and the flight control board through a CAN bus or serial communication interface. The airborne computing unit receives spatial concentration discrete point cloud data collected by the probe, performs topological data analysis, and extracts topological invariants such as the Betti number.

[0041] A wireless communication link is configured between the ground monitoring network subsystem and the airborne detection subsystem. The wireless communication link adopts a point-to-point radio frequency data pass-through module or a 5G mobile cellular network terminal. The wireless communication link is used for bidirectional transmission of associated dimension data and status control commands between the edge computing node of the ground monitoring station and the airborne computing unit of the UAV.

[0042] The flight control panel is responsible for performing basic attitude stabilization and waypoint navigation for the multi-rotor UAV platform.

[0043] Example 2

[0044] See attached document Figure 1-4Based on the UAV remote sensing and ground monitoring collaborative power plant pollution plume tracking system of Embodiment 1, this embodiment provides a method for the ground monitoring network subsystem and the airborne detection subsystem to work collaboratively according to a set time period when performing dynamic tracking tasks of power plant pollution plumes, including the following steps:

[0045] Step S210: The environmental parameter acquisition module of the ground monitoring station acquires the pollutant concentration time series within a preset time window. The edge computing node uses Takens' delayed embedding theorem to reconstruct the pollutant concentration time series into a high-dimensional phase space state vector according to the set time delay parameter and embedding dimension parameter, and extracts the correlation dimension and the maximum Lyapunov exponent of the high-dimensional phase space state vector distribution.

[0046] In step S220, the airborne detection subsystem receives the correlation dimension and the maximum Lyapunov exponent via a wireless communication link. The airborne computing unit generates servo motor drive commands based on the value of the correlation dimension. The servo motor driver adjusts the extension length of the mechanical link of the dynamic variable pitch multidimensional sensor array according to the servo motor drive commands. When the correlation dimension is higher than a set critical threshold, the servo motor driver drives the mechanical link to perform a retraction action; when the correlation dimension is lower than or equal to the set critical threshold, the servo motor driver drives the mechanical link to perform an extension action.

[0047] Step S230: Under the current mechanical link length configuration, the dynamic variable-range multi-dimensional sensing array collects a set of discrete three-dimensional spatial concentration points in the airspace where the UAV is located. The airborne computing unit removes data points whose concentration measurements are lower than the background noise baseline from the set of discrete three-dimensional spatial concentration points, obtaining an effective spatial sampling point set. The airborne computing unit sets a dynamic filtering threshold based on the maximum Lyapunov exponent and uses the dynamic filtering threshold to construct a Vietoris-Rips simplex complex from the effective spatial sampling point set. The formula for calculating the dynamic filtering threshold is as follows:

[0048]

[0049] In the formula, Representing airborne computing power units in The dynamic filtering threshold is calculated at any given time. This represents the system's preset basic spatial connection determination distance; The threshold attenuation coefficient represents the system calibration. Representing edge computing nodes The maximum Lyapunov exponent is calculated at each time step. During the construction of the Vietoris-Rips simplex, the airborne computing unit calculates the Euclidean distance between any two spatial sampling points in the effective spatial sampling point set. If the Euclidean distance between any two spatial sampling points is less than or equal to the dynamic filtering threshold, the airborne computing unit generates a one-dimensional simplex connection between the two spatial sampling points.

[0050] Step S240: The airborne computing unit performs simple homology group algebra calculation on the constructed simple complex and extracts the zero-order Betti number and the first-order Betti number. The zero-order Betti number is used to indicate the number of independent connected branches formed by the effective spatial sampling point set in the current topological space, and the first-order Betti number is used to indicate the number of one-dimensional topological holes formed inside the effective spatial sampling point set.

[0051] In step S250, the airborne computing unit monitors the acquisition status of the dynamic variable-distance multidimensional sensor array in real time. If the concentration measurement values ​​of all probes in the dynamic variable-distance multidimensional sensor array drop below the background noise baseline, the airborne computing unit determines that the multi-rotor UAV platform has entered the clean air fault zone of the pollution plume. The airborne computing unit retrieves the simple complex data of the moment before entering the clean air fault zone and extracts the independent connected branch with the largest number of spatial sampling points as the main connected branch set.

[0052] In step S260, the airborne computing unit calculates the spatial geometric center coordinates of the main connected branch set, constructs the spatial covariance matrix of the main connected branch set based on the spatial geometric center coordinates of the main connected branch set, and extracts the eigenvector corresponding to the largest eigenvalue of the spatial covariance matrix through eigenvalue decomposition. The eigenvector corresponding to the largest eigenvalue is used as the spatial topology main direction vector. The flight control board receives the spatial topology main direction vector and controls the multi-rotor UAV platform to maintain straight-line level flight along the spatial topology main direction vector until the dynamic variable pitch multi-dimensional sensor array re-collects concentration data higher than the background noise baseline.

[0053] In step S270, the airborne computing unit constructs a new simplex using the newly acquired concentration data and calculates the zero-order Betti number of the new simplex. The airborne computing unit extracts the geometric center coordinates of the new simplex and calculates the vertical distance from the geometric center coordinates of the new simplex to the extension line of the principal direction vector of the spatial topology. If the vertical distance is less than the preset coherence matching distance threshold, the airborne computing unit determines that the new simplex and the main connected branch set before crossing the clean air fault area belong to the same pollution plume topology, confirms that the cross-fault tracking is successful, and controls the system to return to step S210 to enter the next spatial sampling and position optimization logic control closed loop.

[0054] Furthermore, the process of time-series data acquisition and phase space reconstruction performed by the ground monitoring network subsystem includes the following steps:

[0055] In step S310, the environmental parameter acquisition module of the ground monitoring station continuously acquires the concentration values ​​of characteristic pollutants according to the set sampling frequency, forming a one-dimensional discrete concentration time series. The edge computing node extracts the one-dimensional discrete concentration time series within a preset time window as the basic data for reconstruction, denoted as the sequence. ,in , This represents the total number of concentration data sampling points included within the preset time window;

[0056] Step S320, edge computing node computation sequence Optimal time delay parameters Edge computing nodes use the mutual information method to solve for the optimal time delay parameters. Specifically, the edge computing nodes compute the sequence... With delayed sequence Mutual information function between Edge computing nodes extract mutual information function The number of delay steps corresponding to the first attainment of a local minimum is used as the optimal time delay parameter. ;

[0057] Step S330, the edge computing node is based on the optimal time delay parameter Calculate the sequence Minimum embedding dimension parameter Edge computing nodes use the spurious nearest neighbor method to solve for the minimum embedding dimension parameter. The edge computing nodes have a dimension... Find the nearest neighbor of a state vector in the phase space, and calculate the value when the dimension increases to... The relative rate of change of the Euclidean distance between the two state vectors mentioned above is calculated using the following formula:

[0058]

[0059] In the formula, Represents in dimension Next The relative rate of change of a state vector after adding one dimension; Representing the The state vector at the th state vector in the th... Dimensional coordinate components; Representing the The nearest neighbor of the nth state vector is at the nth state vector. Dimensional coordinate components; Represents in dimension Next Euclidean distance between each state vector and its nearest neighbor;

[0060] If the relative rate of change exceeds the distance inflation threshold set by the system, the edge computing node determines the aforementioned nearest neighbor as a false nearest neighbor. The edge computing node then calculates the proportion of false nearest neighbors among all state vectors in the phase space. When the proportion of false nearest neighbors decreases to near zero or falls below the set minimum threshold, the edge computing node will adjust the corresponding dimension... Determined as the minimum embedding dimension parameter ;

[0061] Step S340, the edge computing node uses the obtained optimal time delay parameters. With minimum embedding dimension parameter Applying Takens' delayed embedding theorem to one-dimensional discrete concentration time series Phase space reconstruction is performed, and edge computing nodes generate state vectors in a high-dimensional phase space. The formula for constructing the state vector is as follows:

[0062]

[0063] In the formula, The first phase in the high-dimensional phase space obtained by reconstruction A state vector; Represents the first time within the preset time window Individual concentration sampling data; ; This represents the total number of state vectors contained in the high-dimensional phase space. The relationship for calculating the total number of state vectors is as follows: Edge computing nodes use the above formula to map one-dimensional concentration fluctuations on the time scale into high-dimensional geometric evolution trajectories in the topological space, thus establishing the underlying phase space data structure for subsequent extraction of chaotic measures such as correlation dimension.

[0064] Furthermore, based on the aforementioned high-dimensional phase space reconstruction results, the process by which the edge computing nodes of the ground monitoring network subsystem perform correlation dimension calculation and local chaotic feature extraction includes the following steps:

[0065] Step S410: The edge computing node traverses all state vectors in the high-dimensional phase space and calculates the spatial distance between any two state vectors. The spatial distance is rigorously measured using the Euclidean norm. With state vector Edge computing nodes compute state vectors With state vector The L2 norm of the difference vector between the two states is used to obtain the corresponding state vector spacing.

[0066] Step S420: The edge computing nodes are configured with a series of discrete spatial observation scale parameters. And based on the state vector spacing, calculate the correlation integral corresponding to the parameters of each spatial observation scale. The correlation integral is used to statistically determine the proportion of state vector pairs in phase space whose distance is less than a given spatial scale parameter. The specific formula for calculating the correlation integral is as follows:

[0067]

[0068] In the formula, Represents the parameters of a given spatial scale of investigation The related integrals below; This represents the total number of state vectors contained in the high-dimensional phase space; This represents the Heaviside step function. When the input variable is greater than or equal to zero, the value of the Heaviside step function is 1, and when the input variable is less than zero, the value of the Heaviside step function is 0. Represents the state vector With state vector The Euclidean distance between them;

[0069] Step S430: The edge computing nodes construct a discrete data mapping relationship based on double logarithmic coordinates. The edge computing nodes use the natural logarithm of the spatial observation scale parameter. The x-axis represents the natural logarithm of the corresponding correlated integral. Using the vertical axis, scatter data in a double logarithmic coordinate system is generated, and edge computing nodes identify scale-free regions with linear mapping characteristics in the scatter data.

[0070] In step S440, the edge computing node performs least squares linear fitting on the included coordinate scatter data within the defined scale-free region. The edge computing node extracts the slope value of the fitted line and determines the extracted slope value as the correlation dimension of the local flow field. Correlation dimension The mathematical definition and formula are as follows:

[0071]

[0072] In the formula, The correlation dimension represents the local flow field; The natural logarithm of the correlated integral; The natural logarithm, representing the spatial scale parameter, is the correlation dimension calculated by the edge computing nodes. Assign a corresponding timestamp to form a time-varying parameter;

[0073] Step S450, the edge computing node will associate the dimensions with timestamps. The data is transmitted to the airborne detection subsystem, where the onboard computing unit receives the correlation dimension. And assess the nonlinear dynamic state of the contamination plume flow field at the current moment, when the correlation dimension is... As the numerical value increases, the airborne computing unit determines that the current flow field exhibits severe turbulence and increased complexity of its nonlinear structure; when the correlation dimension... When the value decreases, the airborne computing unit determines that the current flow field tends to be in a laminar state and the complexity of the nonlinear structure of the flow field decreases. Based on the above physical judgment results, the airborne computing unit provides the underlying numerical basis for the subsequent hardware linkage of the dynamic variable-span multidimensional sensing array.

[0074] Furthermore, the dynamic variable pitch multidimensional sensor array is mounted at the bottom center of gravity of the multi-rotor UAV platform. The dynamic variable pitch multidimensional sensor array includes a central control shell and four mechanical telescopic links symmetrically distributed around the outer periphery of the central control shell.

[0075] The central control housing integrates a microcontroller, a multi-axis drive circuit board, and a counterweight. The microcontroller is electrically connected to the multi-axis drive circuit board via an I / O interface. The central control housing is made of lightweight carbon fiber material and has a heat dissipation grille at the bottom.

[0076] The mechanical telescopic linkage adopts a three-stage electrically controlled telescopic sleeve structure. Each set of mechanical telescopic linkages is driven by an independent precision stepper motor. The output shaft of the stepper motor is connected to a miniature precision lead screw through a coupling. The miniature precision lead screw cooperates with the lead screw nut pair inside the mechanical telescopic linkage. The rotation of the stepper motor drives the miniature precision lead screw to rotate, driving the mechanical telescopic linkage to perform linear telescopic action in the radial direction.

[0077] Each set of mechanical telescopic linkages is equipped with a probe at its end. The probe housing is made of polytetrafluoroethylene. The probe integrates a semiconductor gas sensor group, a digital temperature sensor, and a miniature air pump. The miniature air pump is connected to the gas sampling port through a gas guide tube.

[0078] The dynamic variable-pitch multidimensional sensor array adjusts the physical spatial baseline distance between each detection probe by changing the extension stroke of the mechanical telescopic link;

[0079] The specific hardware implementation and control process of the dynamic variable-gap multidimensional sensing array includes the following steps:

[0080] In step S510, the microcontroller receives the physical baseline length instruction from the onboard computing unit and converts the physical baseline length instruction into pulse control parameters for each axis stepper motor.

[0081] In step S520, the multi-axis drive circuit board outputs current pulses of a specific frequency to the stepper motor. The stepper motor drives the miniature precision lead screw to rotate, converting the rotational motion into the linear translational motion of the mechanical telescopic link. The real-time physical length of the mechanical telescopic link is calculated using the following formula:

[0082]

[0083] In the formula, Represents the mechanical telescopic linkage in The real-time physical length at any given moment; Represents the initial rest length of the mechanical telescopic link; This represents the total number of pulses executed by the stepper motor from its initial position. The lead represents the lead of a miniature precision ball screw; This represents the inherent number of steps per revolution of a stepper motor; The subdivision number represents the configuration of the multi-axis drive circuit board;

[0084] In step S530, the linear encoder built into each mechanical telescopic link detects the displacement of the link in real time and sends the displacement back to the microcontroller. The microcontroller compares the displacement with the target length and performs closed-loop position correction.

[0085] In step S540, after the mechanical telescopic link reaches the target position, the microcontroller triggers the micro air pump inside the probe to start, the gas sensor group collects the gas concentration electrical signal, and converts it into a digital concentration signal through the analog-to-digital conversion circuit.

[0086] Step S550: The microcontroller acquires the spatial position coordinates and heading angle of the multi-rotor UAV platform in the absolute coordinate system. The microcontroller, combined with the real-time physical length of the mechanical telescopic links, calculates the three-dimensional spatial coordinates of each probe in the absolute coordinate system. The four mechanical telescopic links on the outer periphery of the central control shell are set to be arranged in a cross-shaped vertical distribution on the horizontal plane. The microcontroller then calculates the... The formula for the three-dimensional spatial coordinates of a single probe is as follows:

[0087]

[0088] In the formula, Representing the Each probe is in The absolute three-dimensional spatial coordinates at any given moment; Representing the Multi-rotor Unmanned Aerial Vehicle Platform Center The absolute spatial coordinates at that moment; Representing multi-rotor unmanned aerial vehicle platforms in Horizontal heading angle at any given time; The number represents the probe number, which can be 1, 2, 3 or 4. The microcontroller packages the digital concentration signal of each probe with the calculated three-dimensional spatial coordinates and sends it to the onboard computing unit via the CAN bus.

[0089] Furthermore, the airborne computing unit of the airborne detection subsystem executes a nonlinear mapping driving mechanism of chaotic measure and mechanical scaling based on the received correlation dimension. The specific control closed-loop steps are as follows:

[0090] In step S610, the airborne computing unit receives the correlation dimension with timestamp sent by the edge computing node through the wireless communication link. The digital signal processor in the airborne computing unit extracts the correlation dimension value and stores the correlation dimension value in a preset control register as an environmental variable for the current control cycle.

[0091] In step S620, the airborne computing unit calls the pre-calibrated critical correlation dimension threshold and mechanical response sensitivity coefficient of the system. The system conducts background data tests in areas with no significant pollution sources under historical stable weather conditions to obtain the average correlation dimension under laminar flow conditions. The system adds a set standard deviation multiple constant to the average correlation dimension to calibrate it as the critical correlation dimension threshold. At the same time, the system sets the minimum physical deployment length based on the mechanical limits of the dynamic variable-span multidimensional sensor array and obtains the maximum correlation dimension extremum by combining test data under historical strong turbulence conditions. The system uses the maximum physical deployment length, minimum physical deployment length, maximum correlation dimension extremum, and critical correlation dimension threshold to calculate the mechanical response sensitivity coefficient. The system is based on a first-order exponential decay response model and sets the threshold at which the maximum correlation dimension extremum is reached. At that time, the target physical baseline length exactly decays to the minimum physical deployment length. Therefore, the formula for calculating the mechanical response sensitivity coefficient is derived as follows:

[0092]

[0093] In the formula, Represents the mechanical response sensitivity coefficient; This represents the maximum physical extended length allowed by the mechanical telescopic link; Represents the minimum physical deployment length set by the system; Represents the extreme value of the maximum correlation dimension; This represents the threshold of the critical correlation dimension.

[0094] In step S630, the airborne computing unit extracts the correlation dimension value from the control register and substitutes it into the nonlinear control equation described in step S130 above for solution. The correlation dimension value is mapped to the target physical baseline length of the dynamic variable-distance multidimensional sensing array. The physical logic of this mapping is as follows: when the correlation dimension of the flow field increases, it indicates that the turbulence inside the pollution plume is intense and the distribution of microscopic materials is highly fragmented. At this time, the system shortens the physical baseline length through the control equation to improve the local spatial sampling resolution and capture high-frequency concentration fluctuation characteristics. Conversely, when the correlation dimension decreases (tending to laminar flow), the system extends the physical baseline to expand the macroscopic detection cross-section.

[0095] Step S640: The airborne computing unit performs piecewise response determination based on the mathematical logic of the nonlinear control equations, when the received correlation dimension... Less than or equal to the critical correlation dimension threshold When, the function that takes the maximum value The calculation result is zero, and the target physical baseline length is zero. The calculation result is equal to the maximum physical unfolding length. At this point, the airborne computing unit generates a static maintenance command to control the mechanical telescopic linkage to maintain or extend to the maximum detection cross-section state;

[0096] Step S650, when the received correlation dimension Greater than the critical correlation dimension threshold At that time, the length of the target physical baseline The correlation dimension decreases exponentially with increasing correlation dimension. The airborne computing unit calculates the actual physical length of the current mechanical telescopic link and the target physical baseline length. The airborne computing unit converts the displacement deviation value into the corresponding incremental pulse count and direction level signal of the stepper motor. The formula for calculating the incremental pulse count is as follows:

[0097]

[0098] In the formula, represent The number of incremental pulses sent to the stepper motor by the onboard computing unit at any given time; Represents the length of the target physical baseline; This represents the actual physical length of the current mechanical telescopic linkage; This represents the inherent number of steps per revolution of a stepper motor; The subdivision number represents the configuration of the multi-axis drive circuit board; The lead of a miniature precision ball screw, if the target physical baseline length Smaller than the actual physical length of the current mechanical telescopic linkage The airborne computing unit generates a contraction direction level signal; if the target physical baseline length Greater than the actual physical length of the current mechanical telescopic linkage The airborne computing unit generates a deployment direction level signal;

[0099] In step S660, the onboard computing unit sends the incremental pulse count and direction level signal to the microcontroller inside the central control housing of the dynamic variable-pitch multi-dimensional sensor array via the internal CAN bus. The microcontroller drives the multi-axis drive circuit board to output current pulses of the corresponding frequency to the stepper motor. The stepper motor drives the internal mechanical transmission mechanism to perform the linkage extension and retraction action, so that the spatial baseline distance between the detection probes reaches the target physical baseline length. .

[0100] Furthermore, combining the three-dimensional spatial coordinates and digital concentration signals output by the aforementioned dynamic variable-span multi-dimensional sensing array, the specific steps for the airborne computing unit to perform spatial point cloud acquisition and Vietoris-Rips complex construction are as follows:

[0101] In step S710, within a preset time sliding window, the airborne computing unit continuously receives the digital concentration signals and their corresponding three-dimensional spatial coordinates transmitted from each detection probe. The airborne computing unit compares the digital concentration signals with the system's preset background noise baseline value. The airborne computing unit discards coordinate data whose digital concentration signals are lower than or equal to the background noise baseline value. The airborne computing unit constructs the remaining coordinate data into an effective spatial sampling point set, forming a spatial discrete point cloud. ;

[0102] in Representing the The three-dimensional coordinate vector of each valid sampling point This represents the total number of valid sampling points contained in a spatially discrete point cloud.

[0103] In step S720, the airborne computing unit calculates the dynamic filtering threshold required for simple complex construction based on the maximum Lyapunov exponent synchronously sent by the ground edge computing node. The dynamic filtering threshold is used to determine the topological connectivity between independent data points in the spatial discrete point cloud. The specific calculation formula and parameter definition of the dynamic filtering threshold refer to the calculation formula recorded in step S230 above.

[0104] Step S730, the airborne computing unit calculates the dynamic filtering threshold based on the solution. In spatial discrete point clouds Construct a Vietoris-Rips complex, and denote the constructed Vietoris-Rips complex as . Airborne computing units will discrete point clouds in space Each valid sampling point in Defined as a zero-dimensional simplex, the airborne computing unit traverses the discrete point cloud in space. Calculate any two valid sampling points and The Euclidean distance between them is calculated using the following formula:

[0105]

[0106] In the formula, Represents valid sampling points and The Euclidean distance between them; Represents valid sampling points The three-dimensional spatial coordinate components; Represents valid sampling points The three-dimensional spatial coordinate components, if the Euclidean distance is less than or equal to the dynamic filtering threshold Airborne computing units at effective sampling points and An undirected edge is generated between them, establishing a one-dimensional simplex;

[0107] Step S740: The airborne computing unit expands the simplex determination rules to higher dimensions. To avoid combinatorial explosions and computation timeouts when processing dense point clouds, the system sets a maximum computational coherence dimension. For spatially discrete point clouds any of one (and) The onboard computing unit determines the Euclidean distance between all pairs of points within the extracted subset, which consists of valid sampling points. If the Euclidean distance between any two points within the extracted subset is less than or equal to the dynamic filtering threshold, then the subset is considered valid. The airborne computing unit will... Connecting each valid sampling point establishes a [system / structure]. Multidimensional simplexes. The onboard computing unit aggregates multidimensional simplexes and uses a sparse filtering algorithm to remove redundant higher-order surfaces, thus completing the Vietoris-Rips complex. The onboard computing unit stores the completed Vietoris-Rips complex into memory, providing a definite topological data structure for subsequent algebraic solutions of simple homology groups.

[0108] Furthermore, based on the Vietoris-Rips complex constructed above, the specific steps for the airborne computing unit to perform simple homology group solutions and low-order Betti number extraction are as follows:

[0109] In step S810, the onboard computing unit transforms the Vietoris-Rips complex stored in memory into an algebraic chain group representation. The onboard computing unit traverses the Vietoris-Rips complex and extracts all the... A set of 3D simplexes, with airborne computing power units based on binary fields. For a scalar field, construct a scalar field with A linear space with a basis of a set of 3D simplexes is established as follows: Chainlink Group Statistics of airborne computing units The total number of simplexes contained in the set of dimensional simplexes is recorded as the number of simplexes. ;

[0110] Step S820: The airborne computing unit establishes a boundary homomorphic mapping relationship between chain groups in adjacent dimensions and defines boundary operators. Boundary operators are used to... A simplex is mapped to all elements containing the mapped simplex. Linear combination of dimensional surfaces, in this embodiment, a binary domain is used. Given a basis, addition is equivalent to modulo-2 addition for any given basis. Simplex The boundary operator mapping formula is as follows:

[0111]

[0112] In the formula, Representative effect on Simplex The boundary operator mapping results; Representative by Composed of 10 effective sampling points A sequence of 1D simplex vertices; Representative at Remove the first from the sequence of dimensional surfaces One valid sampling point;

[0113] Step S830, the airborne computing unit is based on the established boundary operators Construct boundary matrices in various dimensions. The airborne computing unit will Chainlink Group The basis simplex in the model is mapped to the column vectors of the boundary matrix, and then... Chainlink Group The basis simplex mapping in the model is represented by row vectors of the boundary matrix. Airborne computing units determine specific... Is the 3D simplex a corresponding simplex? If the boundary of the simplex is determined to be true, the airborne computing unit sets the corresponding element of the boundary matrix to 1; if it is determined to be false, the airborne computing unit sets the corresponding element of the boundary matrix to 0.

[0114] Step S840, the airborne computing unit constructs the boundary matrix. Perform elementary matrix transformations, solve for the rank of the boundary matrix in each dimension, and then... 3D boundary matrix The rank record is , will the 3D boundary matrix The rank record is ;

[0115] Step S850: The onboard computing unit defines the core space as a closed-chain group containing all closed chains. Define the image space as the edge group containing all edge chains. The airborne computing unit calculates the simple homology group based on the rank of the boundary matrix in each dimension. Rank, extract the Betty number of the corresponding dimension. The airborne computing unit solves the rank-zero degree theorem based on the rank-zero degree theorem. Betty number The formula is as follows: In the formula, Representing the Betty number; represent Number of simplex elements; Representing the 3D boundary matrix rank; Representing the 3D boundary matrix rank;

[0116] Step S860: The onboard computing unit extracts the zero-order Betty number. With first-order Betty number The airborne computing unit will use the zero-order Betty number The record represents the number of independent connected components within the current probe airspace. The airborne computing unit will use the first-order Betti number... The number of topological voids formed within the contamination plume is recorded, and the onboard computing unit will use the zero-order Betti number. With first-order Betty number Write to the control register as the basis for subsequent judgments on whether the multi-rotor UAV platform has entered the clean air fault zone and for executing cross-fault trajectory planning.

[0117] Furthermore, combining the aforementioned topological data structure construction and Betty number extraction process, the specific steps taken by the airborne computing unit to perform zero-concentration fault identification and topological manifold principal direction calculation when determining that the multi-rotor UAV platform has entered a clean air fault region are as follows:

[0118] In step S910, the airborne computing unit acquires the digital concentration signals of each probe in the dynamic variable-distance multi-dimensional sensor array in real time. The airborne computing unit continuously compares the digital concentration signals with the background noise baseline value set by the system. When the digital concentration signals of all probes in the dynamic variable-distance multi-dimensional sensor array are continuously lower than or equal to the background noise baseline value within the preset judgment time window, the airborne computing unit triggers the zero-concentration tomography identification mechanism to determine that the multi-rotor UAV platform has entered the clean air tomography region of the polluted plume.

[0119] Step S920: After triggering the zero-concentration fault identification mechanism, the airborne computing unit retrieves the Vietoris-Rips complex data from the control cycle prior to entering the clean air fault region. Based on the zero-order Betti number extracted in the control cycle prior to entering the clean air fault region, the airborne computing unit traverses and counts the independent connected components in the Vietoris-Rips complex. The airborne computing unit extracts the independent connected component containing the most effective spatial sampling points and establishes the independent connected component containing the most effective spatial sampling points as the main connected component set. Let the main connected component set contain... Let the set of spatial sampling points be denoted as _n_. ,

[0120] in Represents the first connected branch set within the main connected branch set. sampling points Three-dimensional spatial coordinate column vector;

[0121] Step S930: The airborne computing unit calculates the main connected branch set. The spatial geometric center coordinates are calculated by taking the arithmetic mean of the column vectors of the three-dimensional spatial coordinates of all sampled points within the main connected component set by the airborne computing unit. The formula for calculating the spatial geometric center coordinates is as follows:

[0122]

[0123] In the formula, The coordinate vector representing the spatial geometric center of the main connected component set; This represents the total number of valid spatial sampling points contained in the main connected component set; Represents the first connected branch set within the main connected branch set. sampling points Three-dimensional spatial coordinate column vector;

[0124] Step S940: Based on the coordinates of the spatial geometric center, the airborne computing unit constructs a third-order spatial covariance matrix of the main connected component set. The spatial covariance matrix is ​​used to characterize the geometric distribution variance of the main connected component set in three-dimensional space and the correlation between variables in different coordinate axis directions. The formula for calculating the third-order spatial covariance matrix is ​​as follows:

[0125]

[0126] In the formula, The third-order spatial covariance matrix representing the main connected component set; This represents the transpose operation of a matrix;

[0127] Step S950, the airborne computing unit constructs the third-order spatial covariance matrix. The airborne computing unit performs eigenvalue decomposition, solves the characteristic equation of the third-order spatial covariance matrix, obtains three non-negative real eigenvalues ​​and their corresponding initial eigenvectors, and extracts the eigenvalue with the largest value from the three non-negative real eigenvalues.

[0128] In step S960, the airborne computing unit extracts the initial feature vector corresponding to the maximum feature value. The airborne computing unit obtains the flight velocity vector of the multi-rotor UAV platform at the moment before entering the clean air fault zone. The airborne computing unit calculates the inner product of the initial feature vector and the flight velocity vector. If the inner product is greater than or equal to zero, the airborne computing unit uses the initial feature vector as the orientation feature vector. If the inner product is less than zero, the airborne computing unit inverts the initial feature vector to generate the orientation feature vector. This step is used to eliminate the 180-degree directional ambiguity of the feature vector in three-dimensional space and ensure the consistency of the tracking direction.

[0129] In step S970, the airborne computing unit performs normalization processing on the orientation feature vector to obtain the spatial topology main direction vector. The airborne computing unit writes the spatial topology main direction vector into the flight guidance command and sends it to the flight control board. The flight control board receives the flight guidance command and controls the multi-rotor UAV platform to maintain straight-line level flight along the spatial topology main direction vector according to the spatial topology main direction vector until the dynamic variable pitch multidimensional sensor array re-acquires a digital density signal higher than the background noise baseline value.

[0130] Furthermore, based on the aforementioned spatial topology principal direction calculation results, the specific steps for the airborne computing unit and flight control board to collaboratively perform cross-fault forced level flight and homology mapping verification when the multi-rotor UAV platform is in a clean air fault region are as follows:

[0131] In step S1010, the flight control board receives the spatial topology principal direction vector from the onboard computing unit. The flight control board converts the spatial topology principal direction vector into a three-dimensional velocity control command for the multi-rotor UAV platform in the absolute coordinate system. The setting of the three-dimensional velocity control command ensures that the multi-rotor UAV platform can still perform physical space displacement along the topological inertial extension trajectory of the pollution plume even in the absence of concentration gradient data guidance. The calculation formula for the three-dimensional velocity control command is as follows:

[0132]

[0133] In the formula, Representing multi-rotor unmanned aerial vehicle platforms in The three-dimensional velocity control command column vector at each moment; Represents the system's preset fault cruise scalar velocity; The main direction vector of the spatial topology after unitization is represented by the flight control board, which drives the underlying rotor motors according to the three-dimensional velocity control commands to control the multi-rotor UAV platform to perform forced straight-line level flight.

[0134] In step S1020, during the forced straight-line level flight maneuver, the airborne computing unit continuously reads the digital concentration signals of each probe in the dynamic variable-pitch multidimensional sensor array at a preset sampling frequency. When the digital concentration signal of a probe in the dynamic variable-pitch multidimensional sensor array is greater than the background noise baseline value set by the system, the airborne computing unit determines that the multi-rotor UAV platform has penetrated the clean air fault area and entered a new polluted area. The airborne computing unit triggers a data interception mechanism and continuously collects the digital concentration signals and three-dimensional spatial coordinates of the probes within a set time sliding window. The airborne computing unit uses the newly acquired three-dimensional spatial coordinates and digital concentration signals to construct a new set of effective spatial sampling points and applies the aforementioned dynamic filtering threshold to generate a new Vietoris-Rips complex. The airborne computing unit calculates the arithmetic mean coordinates of all data points contained in the new Vietoris-Rips complex and extracts them as the geometric center of the new connected component space.

[0135] In step S1030, the airborne computing unit performs spatial geometric verification of the homology mapping. The airborne computing unit calculates the perpendicular distance from the spatial geometric center of the new connected component to the target spatial line. The target spatial line is determined by the spatial geometric center of the main connected component set before the multi-rotor UAV platform enters the clean air fault region and the principal direction vector of the spatial topology. The airborne computing unit uses the L2 norm of the vector cross product to solve for the perpendicular distance from the spatial geometric center of the new connected component to the target spatial line. The formula for calculating the perpendicular distance is as follows:

[0136]

[0137] In the formula, Represents the perpendicular distance from the geometric center of the new connected component space to the straight line in the target space; A three-dimensional coordinate vector representing the geometric center of the newly connected component space; This represents the spatial geometric center vector of the main connected branch set before entering the clean air fault region; Represents the principal direction vector of the spatial topology; Represents the outer product operation of three-dimensional vectors; The Euclidean second norm of the oriented quantity is represented by the number of terms to be found. For a unit vector, the Euclidean 2 norm of the denominator is always 1;

[0138] In step S1040, the airborne computing unit calculates the vertical distance. The numerical value is compared with the system-calibrated coherence matching distance threshold. If the vertical distance is... If the distance is less than or equal to the homology matching threshold, the airborne computing unit performs simple homology group solution on the new Vietoris-Rips complex and extracts the new zero-order Betti number. If the new zero-order Betti number is greater than or equal to 1, the airborne computing unit determines that the new Vietoris-Rips complex satisfies the connectivity condition in the topology. Based on the above numerical comparison results, the airborne computing unit confirms that the new Vietoris-Rips complex and the main connected branch set before entering the clean air fault region belong to the same thermal power plant pollution plume topology, and determines that the cross-fault tracking and homology mapping verification is successful.

[0139] Step S1050: After successful verification of the coherence mapping, the airborne computing unit deactivates the forced straight-line level flight mode and resumes execution of the aforementioned dynamic tracking logic control closed loop based on correlation dimension and topological invariants. If the vertical distance... If the signal exceeds the coherence matching distance threshold, the onboard computing unit determines that the currently captured concentration signal belongs to other interference sources or random noise. The onboard computing unit clears the new Vietoris-Rips complex data and generates a maintenance command to send to the flight control board. The flight control board, in accordance with the maintenance command, controls the multi-rotor UAV platform to continue to perform straight-line level flight along the original spatial topology main direction vector until it captures a pollution plume structure that meets the coherence mapping verification conditions.

Claims

1. A method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring, characterized in that, Includes the following steps: Step S110: The environmental parameter acquisition module of the ground monitoring station acquires the time series of characteristic pollutant concentrations within a preset time window and sends it to the edge computing node of the same station. In step S120, the edge computing node reconstructs the time series of characteristic pollutant concentrations into a high-dimensional phase space state vector, extracts the correlation dimension of the high-dimensional phase space state vector distribution, and sends it to the airborne computing unit of the airborne detection subsystem through a wireless communication link. In step S130, the airborne computing unit substitutes the received correlation dimension into the control equation to convert it into a mechanical displacement command, which is then sent to the servo motor driver of the dynamic variable pitch multidimensional sensor array. The servo motor driver drives the mechanical telescopic link to perform a retracting or extending action according to the mechanical displacement command. Step S140: Under the current mechanical link length configuration, the dynamic variable-distance multi-dimensional sensing array collects a set of discrete points of three-dimensional spatial concentration in the airspace where the UAV is located, and transmits them back to the airborne computing unit via the internal bus. In step S150, the airborne computing unit performs simple homology group algebra calculation on the discrete point set of three-dimensional spatial concentration, extracts the zero-order Betti number and the first-order Betti number, and uses the extracted zero-order Betti number and the first-order Betti number as the basis for determining whether the multi-rotor UAV platform has entered the clean air fault region and for performing cross-fault trajectory planning. The unit then sends flight guidance commands to the flight control board to control the flight of the multi-rotor UAV platform, so as to dynamically adjust the sampling position and continuously update the effective spatial sampling point set, realize continuous cross-fault tracking of the pollution plume from the thermal power plant, and output the spatial topological diffusion trajectory of the pollution plume.

2. The method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring as described in claim 1, characterized in that, In step S120, the specific steps for the edge computing node to reconstruct the high-dimensional phase space state vector include: The optimal time delay parameter of the characteristic pollutant concentration time series is obtained by mutual information method, and the minimum embedding dimension parameter is obtained by spurious nearest neighbor method. Based on the obtained optimal time delay parameter and minimum embedding dimension parameter, the phase space of the characteristic pollutant concentration time series is reconstructed by Takens' delayed embedding theorem to generate a state vector in a high-dimensional phase space.

3. The method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring as described in claim 2, characterized in that, The specific steps for extracting the correlation dimension of the edge computing node include: Traverse all state vectors in the high-dimensional phase space, calculate the Euclidean distance between any two state vectors to obtain the corresponding state vector spacing; set a series of discrete spatial observation scale parameters, and calculate the correlation integral corresponding to each spatial observation scale parameter based on the state vector spacing; Construct a scatter data system with the natural logarithm of the spatial observation scale parameter as the abscissa and the natural logarithm of the corresponding correlation integral as the ordinate, and identify scale-free regions with linear mapping characteristics in the scatter data. Within the defined scale-free region, least squares linear fitting is performed on the included coordinate scatter data, the slope value of the fitted line is extracted, and the extracted slope value is determined as the correlation dimension of the local flow field.

4. The method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring as described in claim 1, characterized in that, In step S130, the governing equation is specifically as follows: In the formula, Represents a dynamic variable pitch multidimensional sensing array in The length of the target physical baseline at any given time; This represents the maximum physical extended length allowed by the mechanical telescopic link; Represents the mechanical response sensitivity coefficient; Representing airborne computing power units in The number of associated dimensions received at any given time; This represents the critical correlation dimension threshold set by the system.

5. The method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring as described in claim 1, characterized in that, In step S150, before the airborne computing unit performs simple homology group algebraic solution on the three-dimensional spatial concentration discrete point set, the Vietoris-Rips complex is constructed through the following steps: The airborne computing unit removes data points whose concentration measurements are lower than or equal to the background noise baseline value from the three-dimensional spatial concentration discrete point set, and constructs the remaining coordinate data into an effective spatial sampling point set. The edge computing node extracts the maximum Lyapunov exponent of the high-dimensional phase space state vector distribution and synchronously sends it to the airborne computing unit; the airborne computing unit calculates the dynamic filtering threshold required for simple complex construction based on the synchronously sent maximum Lyapunov exponent. On the set of effective spatial sampling points, calculate the Euclidean distance between any two effective sampling points. If the Euclidean distance is less than or equal to the dynamic filtering threshold, generate an undirected edge between the two effective sampling points, establish a one-dimensional simplex, and extend the simplex determination rules to higher dimensions to complete the construction of the Vietoris-Rips complex.

6. The method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring as described in claim 5, characterized in that, The airborne computing unit performs algebraic solutions for simple homology groups, and the calculation steps for extracting Betti numbers are as follows: The constructed Vietoris-Rips complex is transformed into an algebraic chain group, and a boundary homomorphic mapping relationship is established between chain groups in adjacent dimensions to construct the boundary matrix of each dimension. Perform elementary matrix transformations on the constructed boundary matrices of each dimension to solve for the rank of each boundary matrix; then, based on the rank-zero diffusivity theorem, solve for the rank of the first boundary matrix. The formula for the order Betti number is as follows: In the formula, Representing the Betty number; represent Number of simplex elements; Representing the The rank of the boundary matrix; Representing the The rank of the boundary matrix.

7. The method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring as described in claim 1, characterized in that, The method also includes a zero-concentration tomography identification mechanism, the specific steps of which are as follows: The airborne computing unit acquires the digital concentration signals of each probe in the dynamic variable-range multi-dimensional sensor array in real time and continuously compares them with the set background noise baseline value. When the digital concentration signals of all probes in the dynamic variable-distance multidimensional sensor array are consistently lower than or equal to the background noise baseline value within the preset judgment time window, it is determined that the multi-rotor UAV platform has entered the clean air gap region of the polluted plume. After triggering the zero-concentration fault identification mechanism, Vietoris-Rips complex data from the previous control cycle before entering the clean air fault region are retrieved. Based on the extracted zero-order Betti number, the independent connected component containing the largest number of effective spatial sampling points is selected and established as the principal connected component set.

8. The method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring as described in claim 7, characterized in that, After establishing the principal connected component set, the airborne computing unit solves for the principal direction vector of the spatial topology, specifically including: Calculate the spatial geometric center coordinates of the main connected component set, and construct the third-order spatial covariance matrix of the main connected component set based on the spatial geometric center coordinates; Perform eigenvalue decomposition on the third-order spatial covariance matrix to extract the eigenvalue with the largest value and its corresponding initial eigenvector; Obtain the flight velocity vector of the multi-rotor UAV platform at the moment before entering the clean air fault region, calculate the inner product of the initial feature vector and the flight velocity vector, if the inner product is less than zero, invert the initial feature vector to generate an oriented feature vector, and normalize the oriented feature vector to obtain the spatial topological principal direction vector.

9. The method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring as described in claim 8, characterized in that, After calculating the principal direction vector of the spatial topology, the method further includes a forced level flight step across fault lines: The flight control board receives the main direction vector of the spatial topology from the airborne computing unit and converts it into a three-dimensional velocity control command for the multi-rotor UAV platform in the absolute coordinate system, thereby controlling the multi-rotor UAV platform to maintain straight-line level flight. During the straight-line level flight maneuver, when the airborne computing unit reads that the digital concentration signal of the probe in the dynamic variable pitch multidimensional sensor array is greater than the background noise baseline value, the data interception mechanism is triggered to continuously collect the digital concentration signal and three-dimensional spatial coordinates of the probe within the set time sliding window. A new set of effective spatial sampling points is constructed using the newly acquired three-dimensional spatial coordinates and digital concentration signals. A new Vietoris-Rips complex is generated by applying a dynamic filtering threshold. The arithmetic mean coordinates of all data points contained in the new Vietoris-Rips complex are calculated and extracted as the geometric center of the new connected component space.

10. The method for tracking pollution plumes from thermal power plants based on the synergy of UAV remote sensing and ground monitoring as described in claim 9, characterized in that, After extracting the spatial geometric center of the new connected component, the airborne computing unit performs spatial geometric verification of the homology mapping, specifically including: The perpendicular distance from the geometric center of the new connected component space to the target space line is solved using the L2 norm of the vector cross product. The target space line is determined by the geometric center of the main connected component set before entering the clean air fault region and the main direction vector of the spatial topology. If the vertical distance is less than or equal to the homology matching distance threshold, perform simple homology group solution on the new Vietoris-Rips complex to extract the new zero-order Betti number. If the new zero-order Betti number is greater than or equal to 1, it is determined that the new Vietoris-Rips complex and the main connected branch set before entering the clean air fault region belong to the same thermal power plant pollution plume topology, confirming the successful verification of cross-fault tracking and coherence mapping, and deactivating the forced straight-line level flight mode.

Images