Information gain-driven observation path planning method for TDLAS three-dimensional gas concentration field reconstruction

Abstract

This invention provides an information gain-driven method for planning observation paths to reconstruct three-dimensional gas concentration fields using UAVs. It employs a TDLAS sensor mounted on a gimbal to acquire laser path integral concentration data during flight via adaptive scanning. The method utilizes the SART algorithm to reconstruct the three-dimensional concentration field online from sparse, irregularly projected data. A grid reliability evaluation model is introduced to quantify reconstruction uncertainty. Optimal observation points are selected based on information gain, and the gimbal attitude is determined, forming a closed loop of "perception-reconstruction-planning" to achieve efficient localization of gas leak sources.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of gas detection and robotic active olfaction, and relates to an information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method, which is used for online reconstruction of gas diffusion concentration field and leakage source location in outdoor dynamic flow field environment. Background Technology

[0002] Natural gas pipeline leaks are frequent, and traditional location methods rely on fixed sensor networks or manual inspections, which suffer from limited coverage or slow response times. Rotary-wing drones, on the other hand, are flexible, adaptable to various environments, and can be rapidly deployed in complex environments. Equipped with gas sensors, they can quickly detect and locate gas leak sources.

[0003] During drone flight, contact-type gas sensors are susceptible to significant errors in gas concentration measurements due to rotor turbulence. To avoid this, highly sensitive, non-contact tunable semiconductor laser absorption spectroscopy (TDLAS) has become the mainstream method for gas detection. However, TDLAS measurements are integrals of gas concentration along the beam path, providing only the integral of concentration over distance and failing to reveal the spatial distribution of gas along the optical path.

[0004] To overcome the spatial distribution gaps in TDLAT path integrals, researchers combined TDLAT with computed tomography (CT) to develop TDLAT (Tunneled Detailed Artificial Intelligence), using algebraic reconstruction techniques (ART) and their derivative algorithms (SART, etc.) to invert two-dimensional or three-dimensional concentration distributions from multi-angle path integral projection data. However, most existing TDLAT studies are based on fixed optical path layouts in laboratories, and validation experiments are limited to stable planar flame furnace environments. Furthermore, complex beam array designs are still limited to fixed measurement devices, failing to address the challenge of reconstructing sparse and irregular projection data on mobile platforms. This has become a bottleneck in extending TDLAT technology to dynamic outdoor environments.

[0005] Active olfaction technology, by simulating the biological olfactory search mechanism through robotic perception and autonomous decision-making, provides a new approach to leak source localization. However, existing methods mostly use gas sensors and anemometers to track gas plumes, paying less attention to the spatial distribution of gases; while TDLAT can provide spatial distribution, it is mostly based on a fixed optical path layout preset in the laboratory and has not yet considered deployment on mobile platforms, resulting in the failure to effectively integrate active olfaction and concentration field reconstruction technologies.

[0006] In summary, the existing technologies have the following technical problems that need to be solved: (1) Data sparsity and irregularity - the projection beams obtained by TDLAS on the UAV mobile platform are sparse and irregularly distributed, with some data redundancy, which traditional reconstruction algorithms cannot adapt to; (2) Lack of perception planning - the existing methods lack a dynamic observation path planning mechanism for the efficiency of concentration field reconstruction, and cannot adaptively select the optimal observation pose according to the degree of information scarcity, resulting in low reconstruction efficiency and poor timeliness of leakage source location. Summary of the Invention

[0007] The present invention aims to overcome the shortcomings of the prior art and provide a method for reconstructing and planning observation paths of three-dimensional gas concentration fields using TDLAS of unmanned aerial vehicles based on information gain. This method can reconstruct the gas diffusion concentration field and locate the leakage source in an outdoor dynamic flow field environment through three-dimensional concentration field reconstruction and information gain-driven perception planning.

[0008] The technical problem solved by this invention is achieved through the following technical solution: An information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method is characterized by the following steps: Step 1) Establishing a regular grid map: Establishing a regular grid map and initializing the grid map parameters; Step 2) 3D Concentration Field Reconstruction: The 3D-DDA algorithm is used to calculate the grid and travel length of each beam, construct the projection matrix W, and use the SART algorithm for iterative reconstruction. After each iteration, non-negative constraints are executed to generate the reconstructed 3D concentration field and update the confidence of each grid. Step 3) Adaptive scanning and data acquisition: Control the gimbal to perform reciprocating scanning with a given yaw center angle as the center, acquire the path integral concentration data of multiple beams, and record the UAV pose and gimbal attitude at the same time. Step 4) Perception-driven observation path planning: Step 4.1 Candidate observation point generation: Based on the current reconstructed 3D concentration field and confidence level, a set of candidate points is generated according to multiple strategies; Step 4.2 Uncertainty Field Calculation: Calculate the uncertainty field of the entire map and set the uncertainty threshold. This is a multiple used to set the global average uncertainty. Step 4.3 Information gain calculation and optimal pose selection: Calculate the expected information gain for each candidate point, select the candidate point with the highest expected information gain as the next target position, and determine the optimal gimbal scanning center angle; Step 5) Drone navigation: Control the drone to fly to the target location of the candidate point with the highest expected information gain, and continuously collect path integral concentration data during the flight; Step 6) Termination condition for leak source location: When the results of multiple scans by the UAV are greater than the average concentration integral, and the standard deviation of the highest concentration point in the reconstructed three-dimensional concentration field is less than the set allowable location error, the leak source location can be determined to be complete; the highest concentration point in the reconstructed three-dimensional concentration field at the final moment is determined as the estimated gas leak source location; otherwise, repeat steps 2), 3), 4), and 5) of the three-dimensional concentration field reconstruction to form a closed loop.

[0009] Moreover, its characteristic is that: in step 1), the initialization of raster map parameters includes resolution and time constant. Observation saturation value Angular diversity saturation value Uncertainty threshold During initialization, all raster concentration values ​​are set to 0, and the confidence level is set to 0.

[0010] Moreover, its characteristic is that the specific steps of step 2) three-dimensional concentration field reconstruction are as follows: The measurement value using the TDLAS sensor is the integral of the gas concentration along the beam path, as shown in equation (1): (1) in For the first Path integral concentration of a beam For the beam path, This is the spatial concentration distribution function; After discretization, the measured values ​​and grid concentrations satisfy a linear relationship, as shown in equation (2): (2) in, , For all A set of measurements for a single beam of light; For the first Gas concentration within each grid cell , For all A set of concentration values ​​within each grid cell; Indicates the first The beam of light in the The length of travel within a grid cell , for The projection matrix; Iterative reconstruction is performed using the SART joint algebraic reconstruction method. In each iteration, for each beam, the error between the currently estimated projection value and the actual measurement value is first calculated, as shown in Equation (3). Then, the error is weighted and distributed back to all the grids through which the beam passes, and the concentration values ​​of these grids are updated, as shown in Equation (4). (3) (4) in This is the relaxation factor, with a value ranging from 0.1 to 0.5. It is used to control the amplitude of each update, prevent oscillations, and control the convergence speed. Indicates the first The sum of the travel lengths of the beams within all grid cells; For the first After the nth iteration calculation The estimated gas concentration within each grid cell; to measure the reliability of the reconstruction results, a grid reliability score is defined. As shown in equation (5): (5) The first item is the time decay factor, which reflects the timeliness of the data; the newer the data, the higher the confidence level. It is a time constant. The first term is the time difference between the current moment and the last observation moment; the second term is the observation count factor, which increases with the number of observations, and tends to saturate after reaching a certain number of observations, to avoid over-reliance on repeatedly observed grids. This represents the saturation value for the number of observations. For the first The third item is the total number of times each grid was observed; the fourth item is the angular diversity factor, which reflects the angular distribution of the beams observing the grid. The greater the angular diversity, the higher the reliability. This represents the saturation value for angular diversity. The size of the set of beam angles for observing this grid.

[0011] Moreover, its characteristic is that: the specific method of adaptive scanning and data acquisition in step 3) is as follows: with the gimbal in a fixed pitch angle, it performs reciprocating scanning within a set range with a given yaw center angle as the center. By analyzing and reconstructing the concentration field, the yaw center angle with the highest concentration is found, and the center of the fan-shaped scanning area is dynamically adjusted to the direction of this yaw center angle; to avoid abrupt changes in the scanning center causing discontinuity in the observation data, a fan-shaped smooth transition mechanism is adopted, assuming the current scanning center is... The target center is The transition process uses linear interpolation as shown in equation (6); (6) in, for The yaw direction is scanned in real time at all times; , is a transition factor that increases linearly with time. The transition factor increases linearly from 0 to 1 with time, so that the scanning center can move smoothly from the current position to the target position. Let be the angle normalization function. The transition process lasts for a set time, during which the scanning center moves smoothly.

[0012] Moreover, its characteristic is that: the specific method for generating the candidate observation point set in step 4.1, based on the current reconstructed three-dimensional concentration field and confidence level, according to multiple strategies, is as follows: The multi-strategy approach is an organic combination of inverse concentration gradient search, backtracking and retry, region utilization, and random exploration strategies to generate a candidate point set. ; Inverse concentration gradient strategy: The current position can be calculated from the reconstructed concentration field. concentration gradient Concentration The vector composed of partial derivatives in the three-dimensional coordinate direction, with the gradient direction being the direction of the fastest increase in concentration, can generate candidate observation points based on the reconstructed concentration field. As shown in equation (7); (7) in, Step size; This is a direction adjustment factor that is dynamically adjusted based on the concentration change trend. Backtracking and retry strategy: When a continuous decrease in concentration is detected, a backtracking mechanism is triggered, causing the drone to return to a historical location with a higher concentration. To avoid getting trapped in local extrema, as shown in equation (8); (8) in, For the backtracking step size; As the backtracking distance factor, it can be taken as... ; Regional utilization strategy: When a high concentration area is detected, use the region centroid... Multiple observation points are generated around a circle with different distance radii and different azimuth angles, centered on the circle. Random exploration strategy: When there is a lack of effective concentration information or all candidate points are excluded, random candidate points are generated.

[0013] Furthermore, the key feature is that: the specific method for step 4.3, information gain calculation and optimal pose selection, is as follows: For each candidate point, the expected information gain is calculated, and the candidate point with the highest expected information gain is selected as the next target position. The method for determining the optimal gimbal scanning center angle is as follows: The calculation of information gain is based on the uncertainty field of gas concentration distribution. The uncertainty field is defined as follows: To measure spatial location An index for the reliability of concentration estimates, quantifying the degree of information scarcity at that location. Spatial location The uncertainty measure of the concentration value can be calculated from the number of observations and the angular diversity for candidate observation points. and gimbal posture ( , Information gain is defined as the weighted optical path length of the high uncertainty region that the observation pose can cover. Finally, multiple sets of pose information gains are calculated for each candidate observation point, and the average value is taken as the expected information gain of the candidate observation point. As shown in equation (9); (9) in, This is a small constant to prevent division by zero; To Number of observations at a location; The angular diversity coefficient represents the observation... The richness of the beam angle distribution at a location, ranging from 0 to 1, indicates that the larger the value, the richer the observation angle, and the more reliable the concentration estimate at that location. It can be defined as being in The ratio of the number of observations from different directions to the saturation threshold; For candidate observation points and gimbal posture ( , Information gain is defined as the weighted optical path length of the high uncertainty region that the observation pose can cover, as shown in equation (10). (10) in Observe the position on the beam path. This represents the uncertainty threshold. For each candidate observation point, calculate the information gain of multiple poses and take the average value as the expected information gain of that point, as shown in Equation (11). (11) in, To preset the number of yaw center angles, To preset the number of pitch angles, For the first A preset yaw center angle, For the first A preset pitch angle, For the point Posture The information gain of a single observation.

[0014] The beneficial effects of this invention are as follows: 1. The TDLAT three-dimensional gas concentration field reconstruction observation path planning method driven by information gain of the present invention overcomes the shortcomings of the existing fixed optical path TDLAT technology and traditional active olfaction method. It adapts to sparse irregular projection data through SART iterative reconstruction, introduces a grid credibility evaluation model to quantify uncertainty, and dynamically plans the observation path with information gain as the criterion to achieve efficient reconstruction of gas diffusion concentration field and leakage source location in outdoor dynamic flow field environment.

[0015] 2. The present invention provides an information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method, which realizes online reconstruction of TDLAS-SART three-dimensional concentration fields. It combines TDLAS path integral measurement with SART iterative reconstruction to solve the problem of three-dimensional reconstruction of sparse and irregular projection data on mobile platforms. The combination of three-dimensional concentration field reconstruction and information gain-driven observation path planning method overcomes the problem of missing spatial distribution of TDLAS path integrals. Based on the online reconstructed three-dimensional concentration field, leakage source localization is performed, which significantly improves the accuracy of leakage source localization.

[0016] 3. The TDLAS three-dimensional gas concentration field reconstruction observation path planning method driven by information gain of this invention adopts a grid credibility evaluation model, which integrates time decay, observation times and angle diversity factors to quantify grid reconstruction uncertainty and provide a decision basis for observation planning. The SART iterative reconstruction algorithm combined with the grid credibility evaluation model effectively processes sparse and irregular projection beam data on the mobile platform, realizing online reconstruction and uncertainty quantification of the three-dimensional concentration field.

[0017] 4. The present invention provides an information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method. The information gain-driven perception planning closed loop defines the information gain by the weighted optical path length of the uncertain field, selects the optimal observation point and determines the gimbal attitude, and guides the UAV to actively observe the area with the least information, forming a "perception-reconstruction-planning" closed loop. The observation point selection method based on information gain guides the UAV to move to the area with the least information, achieving a balance between exploring unknown areas and utilizing known information, and avoiding blind search.

[0018] 5. The TDLAS three-dimensional gas concentration field reconstruction observation path planning method driven by information gain of the present invention does not rely on wind speed and wind direction measurement. It can realize the leakage source location solely based on concentration field reconstruction and information gain driven planning, effectively overcoming the interference of rotor downwash airflow on traditional gas sensors and reducing system complexity.

[0019] 6. The TDLAS three-dimensional gas concentration field reconstruction observation path planning method driven by information gain of this invention integrates multiple strategies, and organically combines inverse concentration gradient, backtracking retry, region utilization and random exploration strategies to ensure the algorithm's adaptability and robustness in each stage of coarse search, fine localization and unknown exploration. Attached Figure Description

[0020] Figure 1 This is a flowchart of the observation path planning method for reconstructing the three-dimensional gas concentration field according to the present invention; Figure 2 This is a flowchart of the perception-driven observation path planning steps in this invention. Figure 3 This is a flight trajectory diagram of an unmanned aerial vehicle (UAV) according to an embodiment of the present invention; Figure 4 This is a concentration field map reconstructed according to an embodiment of the present invention; Figure 5 This is a cross-sectional view of the concentration field in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will now be described in more detail through specific embodiments. These embodiments are intended to provide a further understanding and explanation of the present invention, and are for descriptive purposes only, and do not constitute any limitation on the scope of protection of the present invention.

[0022] An information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method, such as Figure 1 As shown in the flowchart, the method includes the following steps: Step 1) Create a regular grid map: Create a regular grid map and initialize the grid map parameters; In step 1): set the raster parameters, including resolution and time constant. Observation saturation value Angular diversity saturation value Uncertainty threshold During initialization, all raster concentration values ​​are set to 0, and the confidence level is set to 0.

[0023] Step 2) 3D Concentration Field Reconstruction: The 3D-DDA algorithm is used to calculate the grid and travel length of each beam, construct the projection matrix W, and use the SART algorithm for iterative reconstruction. After each iteration, non-negative constraints are executed to generate the reconstructed 3D concentration field and update the confidence of each grid. The measurement value using the TDLAS sensor is the integral of the gas concentration along the beam path, as shown in equation (1): (1) in For the first Path integral concentration of a beam For the beam path, This is the spatial concentration distribution function; After discretization, the measured values ​​and grid concentrations satisfy a linear relationship, as shown in equation (2): (2) in, , For all A set of measurements for a single beam of light; For the first Gas concentration within each grid cell , For all A set of concentration values ​​within each grid cell; Indicates the first The beam of light in the The length of travel within a grid cell , for The projection matrix.

[0024] Iterative reconstruction is performed using the SART joint algebraic reconstruction method. In each iteration, for each beam, the error between the currently estimated projection value and the actual measurement value is first calculated, as shown in Equation (3). Then, the error is weighted and distributed back to all the grids through which the beam passes, and the concentration values ​​of these grids are updated, as shown in Equation (4). (3) (4) in This is the relaxation factor, with a value ranging from 0.1 to 0.5. It is used to control the amplitude of each update, prevent oscillations, and control the convergence speed. Indicates the first The sum of the travel lengths of the beams within all grid cells; For the first After the nth iteration calculation The estimated gas concentration within each grid cell; to measure the reliability of the reconstruction results, a grid reliability score is defined. As shown in equation (5): (5) The first item is the time decay factor, which reflects the timeliness of the data; the newer the data, the higher the confidence level. It is a time constant. The first term is the time difference between the current moment and the last observation moment; the second term is the observation count factor, which increases with the number of observations, and tends to saturate after reaching a certain number of observations, to avoid over-reliance on repeatedly observed grids. This represents the saturation value for the number of observations. For the first The third item is the total number of times each grid was observed; the fourth item is the angular diversity factor, which reflects the angular distribution of the beams observing the grid. The greater the angular diversity, the higher the reliability. This represents the saturation value for angular diversity. The size of the set of beam angles for observing this grid.

[0025] Step 3) Adaptive scanning and data acquisition: Control the gimbal to perform reciprocating scanning with a given yaw center angle as the center, acquire the path integral concentration data of multiple beams, and record the UAV pose and gimbal attitude at the same time. With a fixed pitch angle, the gimbal performs reciprocating scans within a set range, centered on a given yaw center angle. By analyzing and reconstructing the concentration field, the yaw center angle with the highest concentration is identified, and the center of the fan-shaped scanning area is dynamically adjusted to this angle. To avoid abrupt changes in the scanning center that could lead to discontinuities in the observation data, a fan-shaped smooth transition mechanism is employed, assuming the current scanning center is... The target center is The transition process uses linear interpolation as shown in equation (6); (6) in, for The yaw direction is scanned in real time at all times; , is a transition factor that increases linearly with time. The transition factor increases linearly from 0 to 1 with time, so that the scanning center can move smoothly from the current position to the target position. Let be the angle normalization function. The transition process lasts for a set time, during which the scanning center moves smoothly.

[0026] Step 4) Perception-driven observation path planning: 4.1 Candidate observation point generation: Based on the concentration field reconstruction results, a set of candidate observation points is generated according to multiple strategies; The multi-strategy approach is an organic combination of inverse concentration gradient search, backtracking and retry, region utilization, and random exploration strategies to generate a candidate point set. ; Inverse concentration gradient strategy: The current position can be calculated from the reconstructed concentration field. concentration gradient Concentration The vector composed of partial derivatives in the three-dimensional coordinate direction, with the gradient direction being the direction of the fastest increase in concentration, can generate candidate observation points based on the reconstructed concentration field. As shown in equation (7); (7) in, Step size; This is a direction adjustment factor that is dynamically adjusted based on the concentration change trend. Backtracking and retry strategy: When a continuous decrease in concentration is detected, a backtracking mechanism is triggered, causing the drone to return to a historical location with a higher concentration. To avoid getting trapped in local extrema, as shown in equation (8); (8) in, For the backtracking step size; As the backtracking distance factor, it can be taken as... ; Regional utilization strategy: When a high concentration area is detected, use the region centroid... Multiple observation points are generated around a circle with different distance radii and different azimuth angles, centered on the circle. Random exploration strategy: When there is a lack of effective concentration information or all candidate points are excluded, random candidate points are generated.

[0027] 4.2 Calculation of Uncertainty Fields: Spatial location The uncertainty measure of the concentration value can be calculated from the number of observations and the diversity of perspectives, with an uncertainty threshold set as follows. ; 4.3 Information gain calculation and optimal pose selection: Calculate the expected information gain for each candidate point, select the candidate point with the highest expected information gain as the next target position, and determine the optimal gimbal scanning center angle; The specific method for calculating the expected information gain for each candidate point, selecting the candidate point with the highest expected information gain as the next target position, and determining the optimal gimbal scanning center angle is as follows: The calculation of information gain is based on the uncertainty field of gas concentration distribution. The uncertainty field is defined as follows: To measure spatial location An index for the reliability of concentration estimates, quantifying the degree of information scarcity at that location. Spatial location The uncertainty measure of the concentration value can be calculated from the number of observations and the angular diversity for candidate observation points. and gimbal posture ( , Information gain is defined as the weighted optical path length of the high uncertainty region that the observation pose can cover. Finally, multiple sets of pose information gains are calculated for each candidate observation point, and the average value is taken as the expected information gain of the candidate observation point. As shown in equation (9); (9) in, This is a small constant to prevent division by zero; To Number of observations at a location; The angular diversity coefficient represents the observation... The richness of the beam angle distribution at a location, ranging from 0 to 1, indicates that the larger the value, the richer the observation angle, and the more reliable the concentration estimate at that location. It can be defined as being in The ratio of the number of observations from different directions to the saturation threshold; For candidate observation points and gimbal posture ( , Information gain is defined as the weighted optical path length of the high uncertainty region that the observation pose can cover, as shown in equation (10). (10) in To observe the position on the beam path, This represents the uncertainty threshold. Finally, the information gain of multiple poses is calculated for each candidate observation point, and the average value is taken as the expected information gain of that point, as shown in Equation (11). (11) in, To preset the number of yaw center angles, To preset the number of pitch angles, For the first A preset yaw center angle, For the first A preset pitch angle, For the point Posture The information gain of a single observation.

[0028] Finally, the observation point with the highest expected information gain is selected as the next target location, and the yaw center angle of the attitude with the highest information gain at that point is selected as the center angle of the target scanned by the TDLAS gimbal. The specific process is as follows: Figure 2 As shown.

[0029] Step 5) Drone navigation: Control the drone to fly to the target location and continuously collect path integral concentration data during the flight.

[0030] Step 6) Termination condition for leak source location: When the results of multiple scans by the UAV are greater than the average concentration integral, and the standard deviation of the highest concentration point in the reconstructed three-dimensional concentration field is less than the set allowable location error, the leak source location can be determined to be complete; the highest concentration point in the reconstructed three-dimensional concentration field at the final moment is determined as the estimated gas leak source location; otherwise, repeat steps 2), 3), 4), and 5) of the three-dimensional concentration field reconstruction to form a closed loop.

[0031] Method Validation and Results: To validate the proposed method, simulations were conducted using the GADEN gas leak simulation environment and the Hector_quadrotor drone simulation platform. The simulation involved a drone equipped with a TDLAS gas sensor locating a leak source in a natural gas leak simulation environment. The simulation scenario was set as an open environment area of ​​100m × 100m × 80m, with the odor source located at (-9, 10, 0) m. The average wind speed within the area was 2 m / s–3 m / s, and the wind field exhibited significant time-varying characteristics.

[0032] When the UAV takes off from (20,10), it is downwind of the leak source, and the gas spreads downwind in a plume pattern. The distribution of observation points generated by the perception planning exhibits a clear pattern: early observation points are concentrated in the edge region of the plume to quickly determine the plume boundary; mid-term observation points are distributed along the centerline of the plume, gradually approaching the gas leak source; late-term observation points surround the leak source, observing the high-concentration area from multiple angles. The region with the highest information gain is concentrated in the transition zone between the plume center and the edge. These locations can simultaneously cover known high-concentration areas and insufficiently observed neighboring areas, verifying the effectiveness of the information gain strategy. The method relies on inverse concentration gradient, region utilization, and information gain planning to generate observation points covering all directions around the source point, converging to the true source point after multiple iterations. The UAV flight trajectory is as follows: Figure 3 As shown, the UAV flies against the concentration gradient along the plume, effectively approaching the source point. The reconstructed concentration field is as follows: Figure 4 As shown, the reconstructed concentration field clearly reveals that high-concentration areas are concentrated near the leak source, and the concentration field color at the center of the plume structure is closer to red than at the edges. A cross-sectional view of the concentration field is shown below. Figure 5 As shown, the concentration field exhibits significant concentration variations in three-dimensional space, with the color becoming increasingly reddish closer to the leakage source.

[0033] Statistical results from numerous simulations demonstrate that the proposed method exhibits good robustness and accuracy under different initial conditions, with a leakage source location error of no more than 2.0 meters, showing significant improvements in both location accuracy and efficiency.

[0034] Although the embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the scope of the invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. An information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method, characterized in that: The method includes the following steps: Step 1) Establish a regular grid map: Establish a regular grid map and initialize the grid map parameters; Step 2) 3D Concentration Field Reconstruction: The 3D-DDA algorithm is used to calculate the grid and travel length of each beam, construct the projection matrix W, and use the SART algorithm for iterative reconstruction. After each iteration, non-negative constraints are executed to generate the reconstructed 3D concentration field and update the confidence of each grid. Step 3) Adaptive scanning and data acquisition: Control the gimbal to perform reciprocating scanning with a given yaw center angle as the center, acquire the path integral concentration data of multiple beams, and record the UAV pose and gimbal attitude at the same time. Step 4) Perception-driven observation path planning: Step 4.1 Candidate observation point generation: Based on the current reconstructed 3D concentration field and confidence level, a set of candidate points is generated according to multiple strategies; Step 4.2 Uncertainty Field Calculation: Calculate the uncertainty field of the entire map and set the uncertainty threshold. This is a multiple used to set the global average uncertainty. Step 4.3 Information gain calculation and optimal pose selection: Calculate the expected information gain for each candidate point, select the candidate point with the highest expected information gain as the next target position, and determine the optimal gimbal scanning center angle; Step 5) Drone navigation: Control the drone to fly to the target location of the candidate point with the highest expected information gain, and continuously collect path integral concentration data during the flight; Step 6) Termination condition for leak source location: When the results of multiple scans by the UAV are greater than the average concentration integral, and the standard deviation of the highest concentration point in the reconstructed three-dimensional concentration field is less than the set allowable location error, the leak source location can be determined to be complete; the highest concentration point in the reconstructed three-dimensional concentration field at the final moment is determined as the estimated gas leak source location; otherwise, repeat steps 2), 3), 4), and 5) of the three-dimensional concentration field reconstruction to form a closed loop.

2. The information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method according to claim 1, characterized in that: In step 1), the initialization of raster map parameters includes resolution and time constant. Observation saturation value Angular diversity saturation value Uncertainty threshold During initialization, all raster concentration values ​​are set to 0, and the confidence level is set to 0.

3. The information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method according to claim 1, characterized in that: The specific steps for reconstructing the three-dimensional concentration field in step 2) are as follows: The measurement value using the TDLAS sensor is the integral of the gas concentration along the beam path, as shown in equation (1): （1） in For the first Path integral concentration of a beam For the beam path, This is the spatial concentration distribution function; After discretization, the measured values ​​and grid concentrations satisfy a linear relationship, as shown in equation (2): （2） in, , For all A set of measurements for a single beam of light; For the first Gas concentration within each grid cell , For all A set of concentration values ​​within each grid cell; Indicates the first The beam of light in the The length of travel within a grid cell , for The projection matrix; Iterative reconstruction is performed using the SART joint algebraic reconstruction method. In each iteration, for each beam, the error between the currently estimated projection value and the actual measurement value is first calculated, as shown in Equation (3). Then, the error is weighted and distributed back to all the grids through which the beam passes, and the concentration values ​​of these grids are updated, as shown in Equation (4). （3） （4） in This is the relaxation factor, with a value ranging from 0.1 to 0.

5. It is used to control the amplitude of each update, prevent oscillations, and control the convergence speed. Indicates the first The sum of the travel lengths of the beams within all grid cells; For the first After the nth iteration calculation The estimated gas concentration within each grid cell; to measure the reliability of the reconstruction results, a grid reliability score is defined. As shown in equation (5): （5） The first item is the time decay factor, which reflects the timeliness of the data; the newer the data, the higher the confidence level. It is a time constant. The first term is the time difference between the current moment and the last observation moment; the second term is the observation count factor, which increases with the number of observations, and tends to saturate after reaching a certain number of observations, to avoid over-reliance on repeatedly observed grids. This represents the saturation value for the number of observations. For the first The third item is the total number of times each grid was observed; the fourth item is the angular diversity factor, which reflects the angular distribution of the beams observing the grid. The greater the angular diversity, the higher the reliability. This represents the saturation value for angular diversity. The size of the set of beam angles for observing this grid.

4. The information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method according to claim 1, characterized in that: The specific method of adaptive scanning and data acquisition in step 3) is as follows: With the gimbal at a fixed pitch angle, it performs reciprocating scans within a set range, centered on a given yaw center angle. By analyzing and reconstructing the concentration field, the yaw center angle with the highest concentration is identified, and the center of the fan-shaped scanning area is dynamically adjusted to the direction of this yaw center angle. To avoid abrupt changes in the scanning center causing discontinuity in the observation data, a fan-shaped smooth transition mechanism is adopted. Let the current scanning center be... The target center is The transition process uses linear interpolation as shown in equation (6); （6） in, for The yaw direction is scanned in real time at all times; , is a transition factor that increases linearly with time. The transition factor increases linearly from 0 to 1 with time, so that the scanning center can move smoothly from the current position to the target position. Let be the angle normalization function. The transition process lasts for a set time, during which the scanning center moves smoothly.

5. The information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method according to claim 1, characterized in that: Step 4.1, candidate observation point generation, based on the current reconstructed 3D concentration field and confidence level, uses a multi-strategy approach to generate a candidate point set as follows: The multi-strategy approach is an organic combination of inverse concentration gradient search, backtracking and retry, region utilization, and random exploration strategies to generate a candidate point set. ; Inverse concentration gradient strategy: The current position can be calculated from the reconstructed concentration field. concentration gradient Concentration The vector composed of partial derivatives in the three-dimensional coordinate direction, with the gradient direction being the direction of the fastest increase in concentration, can generate candidate observation points based on the reconstructed concentration field. As shown in equation (7); （7） in, Step size; This is a direction adjustment factor that is dynamically adjusted based on the concentration change trend. Backtracking and retry strategy: When a continuous decrease in concentration is detected, a backtracking mechanism is triggered, causing the drone to return to a historical location with a higher concentration. To avoid getting trapped in local extrema, as shown in equation (8); （8） in, For the backtracking step size; As the backtracking distance factor, it can be taken as... ; Regional utilization strategy: When a high concentration area is detected, use the region centroid... Multiple observation points are generated around a circle with different distance radii and different azimuth angles, centered on the circle. Random exploration strategy: When there is a lack of effective concentration information or all candidate points are excluded, random candidate points are generated.

6. The information gain-driven TDLAS three-dimensional gas concentration field reconstruction observation path planning method according to claim 1, characterized in that: Step 4.3, Information Gain Calculation and Optimal Pose Selection: Calculate the expected information gain for each candidate point, select the candidate point with the highest expected information gain as the next target position, and determine the specific method for the optimal gimbal scanning center angle: The calculation of information gain is based on the uncertainty field of gas concentration distribution. The uncertainty field is defined as follows: To measure spatial location An index for the reliability of concentration estimates, quantifying the degree of information scarcity at that location. Spatial location The uncertainty measure of the concentration value can be calculated from the number of observations and the angular diversity for candidate observation points. and gimbal posture ( , Information gain is defined as the weighted optical path length of the high uncertainty region that the observation pose can cover. Finally, multiple sets of pose information gains are calculated for each candidate observation point, and the average value is taken as the expected information gain of the candidate observation point. As shown in equation (9); （9） in, This is a small constant to prevent division by zero; To Number of observations at a location; The angular diversity coefficient represents the observation... The richness of the beam angle distribution at a location, ranging from 0 to 1, indicates that the larger the value, the richer the observation angle, and the more reliable the concentration estimate at that location. It can be defined as being in The ratio of the number of observations from different directions to the saturation threshold; For candidate observation points and gimbal posture ( , Information gain is defined as the weighted optical path length of the high uncertainty region that the observation pose can cover, as shown in equation (10). （10） in To observe the position on the beam path, This represents the uncertainty threshold. For each candidate observation point, calculate the information gain of multiple poses and take the average value as the expected information gain of that point, as shown in Equation (11). （11） in, To preset the number of yaw center angles, To preset the number of pitch angles, For the first A preset yaw center angle, For the first A preset pitch angle, For the point Posture The information gain of a single observation.

Images