Unmanned aerial vehicle cluster dynamic path planning method and system based on Gaussian plume model

Abstract

The invention discloses an unmanned aerial vehicle cluster dynamic path planning method and system based on a Gaussian plume model. The method comprises the steps that chemical plant leakage gas concentration, wind speed data and thermal imaging data are collected; dynamically correcting the diffusion coefficient of the Gaussian plume model according to the wind speed data, and generating a leakage source probability distribution diagram in combination with the leakage gas concentration and the thermal imaging data so as to determine the concentration gradient value and pheromone field intensity of each position of the chemical plant, thereby planning the search path of the unmanned aerial vehicle cluster; and performing grid division on the leakage source probability distribution diagram to determine a search task, performing search task distribution according to an auction algorithm and the real-time state of the unmanned aerial vehicle, performing leakage source search positioning in combination with a search path, and finally marking a leakage point and returning the leakage point to a command center of the unmanned aerial vehicle cluster in real time. Dynamic path planning of the unmanned aerial vehicle cluster is realized in combination with multi-source data fusion, the unmanned aerial vehicles are guided to preferentially search high-probability regions through task allocation, and the positioning precision of leakage sources and the resource utilization rate of the unmanned aerial vehicles are improved.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) control technology, and in particular to a method and system for dynamic path planning of an UAV cluster based on a Gaussian plume model. Background Art

[0002] Hazardous gas leaks in chemical plants pose a major challenge to industrial safety. Traditional leak detection and repair technologies rely on the deployment of fixed sensors for leak monitoring, which often suffers from significant drawbacks such as numerous blind spots and delayed response times. With the advancement of industrial inspection technology, drones equipped with gas sensors are becoming increasingly popular. However, drone inspections often use preset, fixed paths and are unable to adapt to dynamically changing leak scenarios. For example, while dynamic path planning has been proposed for photovoltaic power station inspections, it lacks real-time data support, resulting in a high mission failure rate. Existing inspection tasks not only rely on single data for positioning, but also often suffer from centralization issues in the drone task allocation process. This not only limits positioning accuracy but also wastes drone energy, leaving resource utilization far from optimal. Summary of the Invention

[0003] In order to solve at least one of the technical problems raised above, the present invention provides a method and system for dynamic path planning of a UAV cluster based on a Gaussian plume model.

[0004] In a first aspect, the present invention provides a method for dynamic path planning of a UAV cluster based on a Gaussian plume model, the method comprising:

[0005] Collecting gas leakage concentration, wind speed data, and thermal imaging data from chemical plants through drone swarms;

[0006] The diffusion coefficient of the Gaussian plume model is dynamically modified based on wind speed data, and a leakage source probability distribution map is generated based on the leakage gas concentration, thermal imaging data, and the modified Gaussian plume model.

[0007] Determine the concentration gradient and pheromone field strength at each location in the chemical plant based on the leak source probability distribution map, and plan the search path for the drone cluster based on the concentration gradient and pheromone field strength.

[0008] Gridding the leak source probability distribution map to determine search tasks, allocating search tasks based on the auction algorithm and the real-time status of the drone, and searching and locating the leak source based on the search path and assigned search tasks;

[0009] The leak point is marked based on the leak source location results and transmitted back to the command center of the drone cluster in real time.

[0010] Preferably, the dynamically correcting the diffusion coefficient of the Gaussian plume model according to the wind speed data comprises:

[0011] Compute initial values ​​for the horizontal and vertical diffusion coefficients:

[0012] ;

[0013] ;

[0014] Where, is the horizontal diffusion coefficient, is the vertical diffusion coefficient, is the downwind distance, which indicates the horizontal distance from the leakage source to the current position of the UAV along the wind direction;

[0015] The horizontal diffusion coefficient and the vertical diffusion coefficient are iteratively optimized by the extended Kalman filter.

[0016] Preferably, generating a leakage source probability distribution map based on the leakage gas concentration, thermal imaging data and the modified Gaussian plume model includes:

[0017] Analyze the thermal imaging data and extract the abnormal temperature rise area to generate a binary thermal imaging mask;

[0018] The predicted concentration field is determined using the modified Gaussian plume model, and the leakage source probability distribution map is determined based on the predicted concentration field and Bayesian probability.

[0019] Preferably, determining the leakage source probability distribution map based on the predicted concentration field and Bayesian probability includes:

[0020] Divide the chemical plant area into grids to generate several sub-areas;

[0021] The leakage source location is set as a random variable, and a priori probability is assigned to each sub-region based on the binary thermal imaging mask. The predicted gas concentration of each sub-region is determined based on the predicted concentration field, and the observed temperature is determined based on the thermal imaging data.

[0022] Determine the likelihood function based on the predicted gas concentration and observed temperature, and determine the corresponding posterior probability based on the prior probability and likelihood function of each sub-region; dynamically update the confidence weights of the leaked gas concentration and thermal imaging data to update the posterior probability of each sub-region;

[0023] The leakage source probability distribution map is determined according to the updated posterior probabilities of each sub-region.

[0024] Preferably, after determining the leakage source probability distribution map according to the updated posterior probabilities of the sub-areas, the method further includes:

[0025] Local refinement search is used to improve the grid resolution in each sub-region marked by the binary thermal imaging mask. The posterior probability of each sub-region is optimized according to the gas concentration gradient, and the sequential Monte Carlo method is used to update the leakage source probability distribution map at preset intervals.

[0026] Preferably, planning the search path of the drone cluster according to the concentration gradient value and the pheromone field strength includes:

[0027] Determine the probability of the drone moving:

[0028] ;

[0029] Where, Indicates the drone is moving from its current position Move to adjacent position probability; Represents concentration gradient, reaction position The rate of change of gas concentration; Indicates location The pheromone strength, Indicates that there is adjacent locations.

[0030] Preferably, the gridding of the leakage source probability distribution map to determine the search tasks and the allocation of the search tasks according to the auction algorithm and the real-time status of the drone include:

[0031] Divide the leakage source probability distribution map into several sub-areas by gridding, and mark the leakage probability value for each sub-area, where each sub-area corresponds to a search task;

[0032] Obtain real-time drone status data, including the remaining battery life and the response time required to scan a sub-area;

[0033] Determine the bidding function of the auction algorithm based on the real-time status data of the drone:

[0034] ;

[0035] Where, For drones is the bid price of the sub-region, For drones The remaining power, is the response time of scanning sub-areas, is the leakage probability of the sub-area. The higher the leakage probability, the higher the bid price.

[0036] Search tasks are assigned according to the bidding function, and the task assignment results are optimized with the goal of minimizing total time consumption and balancing energy consumption.

[0037] In a second aspect, the present invention further provides a UAV cluster dynamic path planning system based on a Gaussian plume model, the system comprising:

[0038] A data acquisition unit, used to collect gas leakage concentration, wind speed data, and thermal imaging data from chemical plants through a swarm of drones;

[0039] A probability distribution determination unit is used to dynamically modify the diffusion coefficient of the Gaussian plume model based on wind speed data, and to generate a leakage source probability distribution map based on the leakage gas concentration, thermal imaging data, and the modified Gaussian plume model;

[0040] The search path planning unit is used to determine the concentration gradient value and pheromone field strength at each location of the chemical plant based on the leakage source probability distribution map, and plan the search path of the drone cluster based on the concentration gradient value and pheromone field strength;

[0041] A search task allocation unit is used to grid the leakage source probability distribution map to determine the search tasks, allocate the search tasks according to the auction algorithm and the real-time status of the UAV, and search and locate the leakage source according to the search path and the allocated search tasks;

[0042] The leakage point marking unit is used to mark the leakage point according to the leakage source positioning results and transmit the information back to the command center of the drone cluster in real time.

[0043] In a third aspect, the present invention also provides an electronic device comprising a processor and a memory, wherein the memory is used to store computer program code, and the computer program code comprises computer instructions. When the processor executes the computer instructions, the electronic device executes the method as described in the first aspect above and any possible implementation thereof.

[0044] In a fourth aspect, the present invention also provides a computer-readable storage medium, in which a computer program is stored. The computer program includes program instructions, and when the program instructions are executed by a processor of an electronic device, the processor executes the method as described in the first aspect above and any possible implementation method thereof.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The present invention provides a dynamic path planning method for a drone cluster based on a Gaussian plume model, comprising collecting leakage gas concentration, wind speed data and thermal imaging data of a chemical plant through a drone cluster; dynamically correcting the diffusion coefficient of the Gaussian plume model according to the wind speed data, and generating a leakage source probability distribution map according to the leakage gas concentration, thermal imaging data and the corrected Gaussian plume model; determining the concentration gradient values ​​and pheromone field strengths at various locations in the chemical plant according to the leakage source probability distribution map, and planning a search path for the drone cluster according to the concentration gradient values ​​and the pheromone field strength; gridding the leakage source probability distribution map to determine search tasks, allocating search tasks according to an auction algorithm and the real-time status of the drones, and searching and locating the leakage source according to the search path and the allocated search tasks; marking the leakage point according to the leakage source positioning result, and transmitting the result back to the command center of the drone cluster in real time.

[0047] The present invention dynamically adjusts the diffusion coefficient of the Gaussian plume model based on wind speed data collected in real time by drones. Compared with traditional fixed parameter models, the revised model reduces the diffusion prediction error in areas with complex terrain. By integrating gas concentration, wind speed, and thermal imaging data to generate a probability distribution map of the leakage source, thermal imaging data can identify high-temperature leakage points and complement the concentration data, greatly improving the positioning accuracy. Combining the concentration gradient with the pheromone field intensity, an adaptive search path is generated through an ant colony-gradient hybrid algorithm. The pheromone field simulates the behavior of ant colonies, guiding drones to prioritize searching high-probability areas, which improves search efficiency compared to traditional methods. Based on the auction algorithm and the real-time status of the drone, including the remaining power and response time, tasks are dynamically allocated. The distributed auction mechanism avoids task clustering, thereby improving the success rate of task execution, while improving positioning accuracy while taking into account search efficiency and improving resource utilization.

[0048] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the background technology, the drawings required for use in the embodiments of the present invention or the background technology will be described below.

[0050] The accompanying drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and, together with the specification, are used to explain the technical solutions of the present disclosure.

[0051] Figure 1 A schematic diagram of a flow chart of a method for dynamic path planning of a UAV cluster based on a Gaussian plume model provided by an embodiment of the present invention;

[0052] Figure 2 for Figure 1Schematic diagram of the flow of sub-steps of step S20;

[0053] Figure 3 for Figure 2 Flow chart of the sub-steps of step S202;

[0054] Figure 4 for Figure 1 Schematic diagram of the flow of sub-steps of step S40;

[0055] Figure 5 A structural schematic diagram of a UAV cluster dynamic path planning system based on a Gaussian plume model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0057] The terms "first," "second," and so on, in the description and claims of the present invention and the accompanying drawings are used to distinguish between different objects, not to describe a specific order. Furthermore, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0058] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0059] See also Figure 1 , Figure 1 The following is a flow chart of a method for dynamic path planning of a UAV cluster based on a Gaussian plume model provided by an embodiment of the present invention. Figure 1 As shown, the method includes:

[0060] S10, using drone clusters to collect leaked gas concentration, wind speed data, and thermal imaging data from chemical plants;

[0061] S20, dynamically correcting the diffusion coefficient of the Gaussian plume model based on the wind speed data, and generating a leakage source probability distribution map based on the leakage gas concentration, thermal imaging data, and the corrected Gaussian plume model;

[0062] S30, determining the concentration gradient value and pheromone field strength at each location of the chemical plant based on the leakage source probability distribution map, and planning a search path for the drone cluster based on the concentration gradient value and pheromone field strength;

[0063] S40, gridding the leakage source probability distribution map to determine search tasks, allocating search tasks based on the auction algorithm and the real-time status of the drone, and searching and locating the leakage source based on the search path and the allocated search tasks;

[0064] S50: Mark the leakage point according to the leakage source positioning result, and send it back to the command center of the drone cluster in real time.

[0065] In this embodiment, each drone is equipped with a high-precision gas sensor, such as an infrared gas sensor, to detect leaked gas concentrations. An ultrasonic anemometer is also installed to measure wind speed data in real time. Before the inspection mission begins, the drones are calibrated and debugged to ensure sensor accuracy. The drones are then launched in a cluster and flown over the chemical plant along a pre-defined preliminary inspection route. During flight, the drones continuously collect gas concentration, wind speed, and thermal imaging data, and transmit this data in real time to a ground control center via a built-in communication module, such as a 4G or 5G module.

[0066] In step S20, based on the real-time collected wind speed data, the diffusion coefficient is dynamically corrected using an empirical formula. Combined with the collected leakage gas concentration data and thermal imaging data, and using probability algorithms such as Bayes' theorem, the corrected Gaussian plume model is applied to the entire chemical plant area to calculate the probability of each location being a leakage source, thereby generating a leakage source probability distribution map.

[0067] In step S30, the concentration gradient value of each location is calculated on the generated leakage source probability distribution map. The concentration gradient value reflects the rate of change of gas concentration in space and can be obtained by differentially calculating the gas concentrations at adjacent locations. At the same time, the concept of pheromone field is introduced. Initially, a basic pheromone field is set up on the leakage source probability distribution map. As the drone searches, when a high-probability leakage area is found, the pheromone intensity in that area is increased; and the pheromone will gradually evaporate over time. Based on the concentration gradient value and the pheromone field intensity, a path planning algorithm such as the ant colony algorithm is used to plan the search path for the drone cluster. For example, the drone preferentially searches areas with large concentration gradient values ​​and high pheromone intensity.

[0068] Furthermore, in step S40, the leak source probability distribution map is divided into grids of equal size, with each grid serving as a search task. For example, the entire chemical plant area is divided into 10×10 grids. In the ground control center, an auction algorithm is used to allocate search tasks based on the real-time status of the drones, such as battery level, remaining range, and current location. Each drone acts as a bidder, evaluating and bidding on each search task based on its own status. Based on the bids, the ground control station assigns the search task to the most suitable drone. Following the assigned search task and planned search path, the drone conducts a detailed search of the corresponding grid area. During the search process, data is continuously collected and the leak source probability distribution map and pheromone field are updated.

[0069] Finally, once the drone locates the leak source during the search, it marks the leak point on the leak source probability distribution map. Different colors or symbols can be used to highlight the leak point. The drone uses a communication module to transmit the leak point location information, related gas concentration data, thermal imaging images, and other information back to the drone cluster's command center in real time. Based on this information, the command center can promptly implement appropriate emergency measures, such as evacuating personnel and activating leak treatment equipment.

[0070] This embodiment uses multi-source data fusion—combining gas concentration, wind speed, and thermal imaging data with a dynamically modified Gaussian plume model—to more accurately locate leak sources, reducing misjudgments and missed detections. Using concentration gradients and pheromone field strength to plan search paths and assigning tasks through an auction algorithm enables drone swarms to conduct searches more efficiently, shortening the time required to locate leak sources. Drones can transmit real-time leak point information to a command center, enabling timely decisions and effective emergency response measures to mitigate the impact of chemical plant leaks on personnel and the environment.

[0071] In one embodiment, dynamically correcting the diffusion coefficient of the Gaussian plume model based on wind speed data includes:

[0072] Compute initial values ​​for the horizontal and vertical diffusion coefficients:

[0073] ;

[0074] ;

[0075] Where, is the horizontal diffusion coefficient, is the vertical diffusion coefficient, is the downwind distance, which indicates the horizontal distance from the leakage source to the current position of the UAV along the wind direction;

[0076] The horizontal diffusion coefficient and the vertical diffusion coefficient are iteratively optimized by the extended Kalman filter.

[0077] First, the various parameters of this embodiment are described:

[0078] : Downwind distance, in meters, represents the horizontal distance from the leakage source to the current position of the UAV along the wind direction, which determines the spatial dependence of the diffusion coefficient;

[0079] : horizontal diffusion coefficient, in meters, describing the diffusion range of gas in the horizontal direction (perpendicular to the wind direction);

[0080] : vertical diffusion coefficient, in meters, describes the diffusion range of gas in the vertical direction;

[0081] : Real-time wind speed, in meters per second, measured by the wind speed sensor carried by the drone, which directly affects the gas diffusion rate;

[0082] : turbulence intensity (dimensionless), defined as the standard deviation of wind speed fluctuations and average wind speed The ratio of , characterizing the strength of atmospheric turbulence.

[0083] Specifically, the above formula is an empirical formula for calculating the diffusion coefficient, where:

[0084] represents the initial diffusion rate, indicating that the diffusion coefficient increases linearly with distance at close distances;

[0085] It is used to correct long-distance diffusion and inhibit the rapid growth of diffusion rate, which conforms to the actual diffusion law.

[0086] represents the vertical diffusion rate, which is lower than the horizontal diffusion rate (limited by gravity and atmospheric stability);

[0087] For controlling the saturation effect of vertical diffusion, the coefficient adjustment is more significant (0.0015 > 0.0001), reflecting that vertical diffusion tends to stabilize more quickly.

[0088] When dynamically correcting the diffusion coefficient, the process is as follows:

[0089] Step 1), initial value calculation:

[0090] According to real-time wind speed and downwind distance , substitute into the above formula to calculate and The initial estimate of .

[0091] Step 2), Extended Kalman Filter (EKF) iterative optimization:

[0092] State variables: , ;

[0093] Observed variables: UAV-measured gas concentrations ;

[0094] Prediction step: Use the current , Substitute into the Gaussian plume model to predict the concentration :

[0095] ;

[0096] Where, is the process noise covariance matrix, is the horizontal distance, is the vertical distance;

[0097] Step 3), iterative diffusion coefficient:

[0098] 3.1) Calculate residuals ;

[0099] 3.2) Through Kalman gain Adjust the diffusion coefficient:

[0100] ;

[0101] 3.3) Update the error covariance matrix to prepare for the next iteration.

[0102] 3.4) Correcting process noise: turbulence intensity The process noise covariance matrix that affects the Kalman filter is When it is larger, the diffusion coefficient changes dramatically and needs to be increased To improve the filter's response speed to observation. The formula is:

[0103] ;

[0104] Where, is the baseline noise intensity, calibrated through experiments.

[0105] In an example scenario, suppose the drone is downwind from a chemical plant. The wind speed was measured at , turbulence intensity , and substituting it into the calculation formula of the diffusion coefficient, we get: , ;

[0106] If the measured concentration , model prediction , then the residual ;

[0107] By Kalman gain After adjustment, we get , , making the predictions closer to the observed values.

[0108] Therefore, the dynamic correction method for the diffusion coefficient of this embodiment achieves real-time adaptation of the Gaussian plume model parameters through empirical formula initialization and Kalman filter iterative optimization, significantly improving the accuracy and robustness of leakage source tracking. The diffusion coefficient is adjusted by real-time wind speed and turbulence intensity to overcome the errors of traditional static models. By extending the Kalman filter to fuse measured concentration data and iteratively optimize model parameters, it is suitable for complex terrain and variable wind speed scenarios in chemical plants, improving calculation accuracy. At the same time, an empirical formula is used to provide a reasonable initial value, reducing the number of Kalman filter iterations, making it suitable for deployment in drone embedded systems, reducing the amount of calculation and improving calculation efficiency.

[0109] See also Figure 2-3 , Figure 2-3 As shown in the figure, in one embodiment, generating a leakage source probability distribution map based on the leakage gas concentration, thermal imaging data and the modified Gaussian plume model includes:

[0110] S201, parsing the thermal imaging data, extracting the abnormal temperature rise area to generate a binary thermal imaging mask;

[0111] S202. Determine a predicted concentration field using the modified Gaussian plume model, and determine a leakage source probability distribution map based on the predicted concentration field and Bayesian probability.

[0112] Furthermore, determining the leakage source probability distribution map based on the predicted concentration field and Bayesian probability includes:

[0113] S2021. Divide the chemical plant area into grids to generate several sub-areas;

[0114] S2022. Set the leak source location as a random variable, assign a priori probability to each sub-region based on the binary thermal imaging mask, determine the predicted gas concentration of each sub-region based on the predicted concentration field, and determine the observed temperature based on the thermal imaging data;

[0115] S2023. Determine a likelihood function based on the predicted gas concentration and the observed temperature, and determine a corresponding posterior probability based on the prior probability and the likelihood function for each sub-region; dynamically update the confidence weights of the leaked gas concentration and the thermal imaging data to update the posterior probability for each sub-region;

[0116] S2024. Determine a leakage source probability distribution map based on the updated posterior probabilities of each sub-region.

[0117] In this embodiment, in order to obtain an accurate leakage-source probability distribution map, the following steps are included:

[0118] Step 1: Data preprocessing:

[0119] 1.1) Process the collected data. The leakage gas concentration is usually collected using a sensor. Suppose the collected original concentration data is , through temperature compensation and cross-interference correction, such as humidity interference on H2S sensor, the calibrated concentration value is :

[0120] ;

[0121] Where, is the humidity correction factor, is the current humidity, To calibrate humidity;

[0122] 1.2) Analyze the thermal imaging data and extract the abnormal temperature rise area to generate a binary thermal imaging mask. This includes acquiring thermal signals through an infrared thermal imager, extracting the abnormal temperature rise area, such as the area where the temperature is more than 5°C higher than the ambient temperature, and generating a binary thermal imaging mask. , used to eliminate non-leaking heat sources (such as steam pipes);

[0123] 1.3) Align the calibrated gas concentration data and thermal imaging data with the GPS location and timestamp to construct a spatiotemporally unified data cube to facilitate subsequent calculations.

[0124] Step 2: Concentration field prediction:

[0125] Use current , Substitute into the Gaussian plume model to predict the concentration :

[0126] ;

[0127] Where, is the process noise covariance matrix, which characterizes the leakage source strength, where The specific size can be determined by least squares inversion.

[0128] Step 3: Bayesian probability fusion and update:

[0129] 3.1) Locate the leak source Set as a random variable based on the thermal imaging mask Assign prior probability ,Preferably, the prior probability of abnormal temperature rise area is 80%, and that of other areas is 20%;

[0130] ;

[0131] Where, Represents the posterior probability, which means that when the observed data is known Under the condition of The probability of Represents the likelihood function, assuming that the leakage source is located at When the data is observed The probability of It means "proportional".

[0132] 3.2) Determine the likelihood function based on the predicted gas concentration and observed temperature:

[0133] ;

[0134] Where, Indicates the degree of match between the observed concentration and the model-predicted concentration, Indicates the degree of match between the thermal imaging temperature distribution and the model prediction, Represents the noise variance of the gas concentration sensor, reflecting the uncertainty of gas concentration measurement. It represents the noise variance of the thermal imaging sensor and reflects the uncertainty of temperature measurement; represents the Euclidean distance, which is used to calculate the global difference between the observed temperature and the predicted temperature distribution; represents the temperature distribution observed in the thermal imaging data, Represents the temperature distribution predicted by the thermal radiation model of the leak source.

[0135] Specifically, The target area can be scanned by an infrared thermal imager to obtain the original thermal radiation signal, and then the thermal radiation intensity can be converted into a temperature value according to the Stefan-Boltzmann law:

[0136] ;

[0137] Where, is the thermal radiation intensity; The emissivity of the surface of the object needs to be preset according to the material, such as the emissivity of the metal pipe. About 0.8, is the Stefan-Boltzmann constant, which is .

[0138] It is a temperature field predicted by a mathematical model based on the thermodynamic characteristics of the leak source, used to describe the thermal radiation effect caused by the leak. For relatively small-scale leaks in chemical plants, the point source model is usually used:

[0139] ;

[0140] Where, is the ambient background temperature, measured by the thermal imager in the non-leak area; is the thermal power of the leakage source, which is positively correlated with the enthalpy and flow rate of the leaked material; is the atmospheric heat attenuation coefficient, which is related to humidity and wind speed; is the distance from the current position to the leak source coordinates;

[0141] in, ;

[0142] Where, is humidity, unit is %. is the wind speed in m / s.

[0143] It should be noted that in the above formula, The smaller The closer it is to 1, the closer it is to 0. The smaller the value, the smaller the temperature distribution difference. The closer it is to 1, the closer it is to 0.

[0144] 3.3) Dynamically update the confidence weights of the leaked gas concentration and thermal imaging data to update the posterior probability of each sub-region:

[0145] Assume that the confidence weights of gas concentration and thermal imaging data are 、 , and the posterior probability is described as:

[0146] ;

[0147] In this formula, 、 The initial value of is 1, and the confidence weight of the data is determined by the sensor that collects the data, which is used to dynamically balance the contribution of gas concentration and thermal imaging data to the posterior probability.

[0148] Preferably, the adjustment rule can be: if a sensor data is continuously abnormal, for example, the concentration drops suddenly but the thermal image does not change, then reduce its weight. The weight is reduced from 1 to 0.7, that is, the weight is 1 in normal state and decays exponentially in abnormal state. In another embodiment, when in a high temperature and high pressure leakage scenario, the weight is increased. and reduce the value of On the contrary, if it is a low temperature leak scenario, it tends to rely on gas concentration data, then it is necessary to increase value, reduce The value of .

[0149] In a preferred embodiment, after determining the leakage source probability distribution map according to the updated posterior probabilities of the sub-regions, the method further includes:

[0150] Local refinement search is used to improve the grid resolution in each sub-region marked by the binary thermal imaging mask. The posterior probability of each sub-region is optimized according to the gas concentration gradient, and the sequential Monte Carlo method is used to update the leakage source probability distribution map at preset intervals.

[0151] In this embodiment, the factory area can be divided into 1m×1m grids, and the posterior probability of each grid point is calculated to generate an initial probability map. Local refinement with a resolution of 0.2m is performed in the heat source mask mark area, and the probability distribution is optimized in combination with the gas concentration gradient ∇𝐶. Then, the probability map is updated by the sequential Monte Carlo method at preset intervals, for example, every 5 seconds. Specifically, particles are densely scattered in high-confidence areas, and particle weights are adjusted and resampled according to new observation data. Finally, the first three local maxima in the probability map are extracted as candidate points of the leakage source. Finally, the thermal probability map is visualized and the coordinates of the candidate points are output to the drone cluster to perform the search task.

[0152] In the above implementation, temperature compensation and humidity correction of gas concentrations were used to eliminate environmental interference and improve the reliability of sensor data. A binary mask was generated based on areas of abnormal temperature rise to eliminate interference from non-leakage heat sources such as steam pipes and narrow the search range. By combining the joint likelihood function of gas concentration and thermal imaging, and matching the observed data with a Gaussian plume model and a point source thermal radiation model, the false alarm rate was significantly reduced. Furthermore, weights were dynamically adjusted based on sensor data anomalies, and Sequential Monte Carlo updates were used to integrate new data and correct probability distributions in real time, suppressing short-term environmental noise and improving the stability and accuracy of positioning results. The coordinates of the top three leak source candidates were output to the drone cluster, prioritizing the search of high-probability areas. This significantly shortened the search time compared to a full area traversal.

[0153] Therefore, this embodiment achieves a triple breakthrough in the accuracy, speed and robustness of leakage source positioning through multi-source data collaborative correction, Bayesian probability dynamic fusion, adaptive resource allocation and real-time visual feedback, significantly improving the level of industrial safety protection.

[0154] In one embodiment, planning a search path for a drone cluster based on the concentration gradient value and the pheromone field strength includes:

[0155] Determine the probability of the drone moving:

[0156] ;

[0157] Where, Indicates the drone is moving from its current position Move to adjacent position probability; Represents concentration gradient, reaction position The rate of change of gas concentration; Indicates location The pheromone strength, Indicates that there is adjacent locations.

[0158] In the chemical plant leak source tracking scenario, the movement probability Indicates the drone is moving from its current position Move to adjacent position The probability of this is determined by two core factors, namely the concentration gradient , reaction position The greater the gradient of the gas concentration change rate, the closer it is to the leak source; the pheromone intensity , characterizing the position of the drone swarm The exploration history of the region is shown in Figure 2. The lower the intensity, the less times the region has been explored. By using the ant colony-gradient hybrid algorithm, a global search can be performed to avoid falling into the local optimum. This means giving gradients a higher weight and moving preferentially to high-concentration areas. At the same time, pheromones are used to suppress repeated paths. For example, if the pheromones in the searched high-gradient area are enhanced, the subsequent drones will reduce their selection probability. After that, it can be used as the input for search task allocation, combined with the UAV power and efficiency constraints to achieve cluster synergy.

[0159] In an example, suppose a drone is located in a chemical plant area. , the adjacent position can be 、 、 , the position parameters are as follows:

[0160] Table 1 The probability of a drone moving to different adjacent locations ;

[0161] As can be seen from the table above, The probability of moving is the largest, so the drone will choose the one with the highest probability As the next target, if another drone arrives , its pheromone intensity Will be updated to a lower value to guide other drones to reduce repeated searches of the area.

[0162] Therefore, this hybrid algorithm, through the collaboration of gradient guidance and pheromones, achieves efficient localization of leak sources in the complex environment of a chemical plant by prioritizing gradient weights and rapidly approaching them. Pheromones avoid repeated paths, adapt to dynamic diffusion scenarios, and improve robustness. It also supports multi-machine collaboration and improves global search efficiency through probability allocation.

[0163] See also Figure 4 In one embodiment, the gridding of the leakage source probability distribution map to determine the search tasks and the allocation of the search tasks according to the auction algorithm and the real-time status of the drone include:

[0164] S401, gridding the leakage source probability distribution map to generate several sub-areas, marking each sub-area with a leakage probability value, wherein each sub-area corresponds to a search task;

[0165] S402: Acquire real-time status data of the drone, including the remaining battery power of the drone and the response time required to scan a sub-area;

[0166] S403: Determine the bidding function of the auction algorithm based on the real-time status data of the drone:

[0167] ;

[0168] Where, For drones is the bid price of the sub-region, For drones The remaining power, is the response time of scanning sub-areas, is the leakage probability of the sub-area. The higher the leakage probability, the higher the bid price.

[0169] S404: Search task allocation is performed according to the bidding function, and the task allocation result is optimized with the goal of minimizing the total time consumption and balancing the energy consumption.

[0170] In this embodiment, in the chemical plant leak emergency inspection scenario, through a dynamic task allocation mechanism (auction algorithm), areas with high possibility of leakage sources, that is, high confidence areas, are reasonably allocated to a cluster system composed of multiple drones to achieve efficient collaborative search.

[0171] First, divide the probability distribution map into several sub-areas based on a spatial grid. For example, divide the factory area into a 10×10 grid, with each grid corresponding to a subtask and each sub-area. Then, mark the leakage probability value. For example, a probability greater than 80% is a high confidence area.

[0172] During inspections, drones report their status in real time, including remaining battery life, sensor response time, and location. The remaining battery life determines the duration of the mission, while the sensor response time indicates how long it takes to scan a subarea, depending on the sensor type and environmental interference. The drone's location determines the energy consumption and time required to reach the target area.

[0173] Furthermore, the auction is conducted according to the bidding function in step S403, and the process is as follows:

[0174] a) The master node (or pilot drone) publishes all sub-area tasks;

[0175] b) Each drone calculates the bid value for each sub-area based on its own status;

[0176] c. The master control node selects the UAV-task pair with the highest bid value and assigns the task;

[0177] d Update the remaining task list and repeat until all tasks are assigned.

[0178] Finally, in step S404, the task allocation optimization goal is determined, including:

[0179] Minimize total time consumption: prioritize assigning high-probability areas to drones with short response times;

[0180] Balanced energy consumption: Avoid single drones running out of power and dynamically adjust task allocation;

[0181] Avoiding duplicate coverage: The searched area is marked by the pheromone field, and the drone avoids high pheromone areas.

[0182] If the battery level of a drone falls below a threshold (e.g., 20%), its unfinished mission will re-enter the auction pool; when a new high-confidence area is discovered, a new round of auction will be triggered.

[0183] For example, in an exemplary scenario, assume that three drones, UAV1, UAV2, and UAV3, participate in task allocation:

[0184] Sub-area A: Leakage probability 90%, scanning time 5 minutes;

[0185] Sub-area B: Leakage probability 85%, scanning time 8 minutes;

[0186] Drone status:

[0187] UAV1 has 80% remaining battery and the scanning speed is fast ( minute);

[0188] UAV2 has 60% remaining battery and the scanning speed is medium ( minute);

[0189] UAV3 has 40% remaining battery and the scanning speed is slow ( minute).

[0190] The bid is calculated as follows:

[0191] UAV1's bid value for sub-area A:

[0192] ;

[0193] UAV2's bid value for sub-area A:

[0194] ;

[0195] UAV3 may give up bidding for high-time-consuming areas due to insufficient battery power (40%).

[0196] Therefore, the final allocation results are as follows: UAV1 is assigned to search sub-area A, UAV2 is assigned to search sub-area B, and UAV3 is on standby or performing low-priority tasks.

[0197] This task allocation method responds to real-time changes in drone status and environmental updates, maximizes overall swarm efficiency through a bidding mechanism, and supports task interruption and reallocation. This solution enables multi-drone swarms to efficiently and reliably complete leak source tracking tasks in complex chemical plant environments.

[0198] In summary, the method provided by the present invention dynamically adjusts the diffusion coefficient of the Gaussian plume model based on wind speed data collected in real time by drones. Compared with traditional fixed parameter models, the revised model reduces the diffusion prediction error in areas with complex terrain; by integrating gas concentration, wind speed, and thermal imaging data to generate a probability distribution map of the leakage source, thermal imaging data can identify high-temperature leakage points, and complements the concentration data, greatly improving the positioning accuracy; combining the concentration gradient and pheromone field intensity, an ant colony-gradient hybrid algorithm is used to generate an adaptive search path. The pheromone field simulates the behavior of ant colonies, guiding drones to prioritize searching high-probability areas, which improves search efficiency compared to traditional methods. Based on the auction algorithm and the real-time status of the drone, including the remaining power and response time, tasks are dynamically allocated. The distributed auction mechanism avoids task clustering, thereby improving the success rate of task execution, while improving positioning accuracy while taking into account search efficiency and improving resource utilization.

[0199] See also Figure 5In one embodiment, the present invention further provides a UAV cluster dynamic path planning system based on a Gaussian plume model, the system comprising:

[0200] The data acquisition unit 100 is used to collect gas leakage concentration, wind speed data and thermal imaging data from the chemical plant through a drone cluster;

[0201] The probability distribution determination unit 200 is used to dynamically modify the diffusion coefficient of the Gaussian plume model according to the wind speed data, and generate a leakage source probability distribution map according to the leakage gas concentration, thermal imaging data, and the modified Gaussian plume model;

[0202] The search path planning unit 300 is used to determine the concentration gradient value and pheromone field strength at each location of the chemical plant based on the leakage source probability distribution map, and plan the search path of the drone cluster based on the concentration gradient value and pheromone field strength;

[0203] The search task assignment unit 400 is used to grid the leakage source probability distribution map to determine the search tasks, assign the search tasks according to the auction algorithm and the real-time status of the UAV, and search and locate the leakage source according to the search path and the assigned search tasks;

[0204] The leakage point marking unit 500 is used to mark the leakage point according to the leakage source positioning result and transmit it back to the command center of the drone cluster in real time.

[0205] It can be understood that the functions or modules included in the system provided in this embodiment can be used to execute the method described in the above method embodiment. Its specific implementation can refer to the description of the above method embodiment. For the sake of brevity, it will not be repeated here.

[0206] The present invention also provides an electronic device, including a processor and a memory, wherein the memory is used to store computer program code, and the computer program code includes computer instructions. When the processor executes the computer instructions, the electronic device executes a method as described in any one of the possible implementation modes.

[0207] The present invention also provides a computer-readable storage medium, in which a computer program is stored. The computer program includes program instructions. When the program instructions are executed by a processor of an electronic device, the processor executes a method as described in any one of the possible implementation methods described above.

[0208] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0209] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here. Those skilled in the art will also clearly understand that the descriptions of the various embodiments of the present invention have different focuses. For the convenience and brevity of description, the same or similar parts may not be repeated in different embodiments. Therefore, for parts not described or not described in detail in one embodiment, reference can be made to the descriptions of other embodiments.

Claims

1. A method for dynamic path planning of UAV clusters based on Gaussian plume model, characterized in that: The method comprises: Collecting gas leakage concentration, wind speed data, and thermal imaging data from chemical plants through drone swarms; The diffusion coefficient of the Gaussian plume model is dynamically modified based on wind speed data, and a leakage source probability distribution map is generated based on the leakage gas concentration, thermal imaging data, and the modified Gaussian plume model. Determine the concentration gradient and pheromone field strength at each location in the chemical plant based on the leak source probability distribution map, and plan the search path for the drone cluster based on the concentration gradient and pheromone field strength. Gridding the leak source probability distribution map to determine search tasks, allocating search tasks based on the auction algorithm and the real-time status of the drone, and searching and locating the leak source based on the search path and assigned search tasks; The leak point is marked based on the leak source location results and transmitted back to the command center of the drone cluster in real time.

2. The method for dynamic path planning of a UAV cluster based on a Gaussian plume model according to claim 1, characterized in that: The method of dynamically correcting the diffusion coefficient of the Gaussian plume model according to wind speed data includes: Compute initial values ​​for the horizontal and vertical diffusion coefficients: ； ； Where, is the horizontal diffusion coefficient, is the vertical diffusion coefficient, is the downwind distance, which indicates the horizontal distance from the leakage source to the current position of the UAV along the wind direction; The horizontal diffusion coefficient and the vertical diffusion coefficient are iteratively optimized by the extended Kalman filter.

3. The method for dynamic path planning of a UAV cluster based on a Gaussian plume model according to claim 1, characterized in that: The method of generating a leakage source probability distribution map based on the leakage gas concentration, thermal imaging data, and the modified Gaussian plume model includes: Analyze the thermal imaging data and extract the abnormal temperature rise area to generate a binary thermal imaging mask; The predicted concentration field is determined using the modified Gaussian plume model, and the leakage source probability distribution map is determined based on the predicted concentration field and Bayesian probability.

4. The method for dynamic path planning of a UAV cluster based on a Gaussian plume model according to claim 3, characterized in that: Determining the leakage source probability distribution map according to the predicted concentration field and Bayesian probability includes: Divide the chemical plant area into grids to generate several sub-areas; The leakage source location is set as a random variable, and a priori probability is assigned to each sub-region based on the binary thermal imaging mask. The predicted gas concentration of each sub-region is determined based on the predicted concentration field, and the observed temperature is determined based on the thermal imaging data. Determine the likelihood function based on the predicted gas concentration and observed temperature, and determine the corresponding posterior probability based on the prior probability and likelihood function of each sub-region; dynamically update the confidence weights of the leaked gas concentration and thermal imaging data to update the posterior probability of each sub-region; The leakage source probability distribution map is determined according to the updated posterior probabilities of each sub-region.

5. The method for dynamic path planning of a UAV cluster based on a Gaussian plume model according to claim 4, characterized in that: After determining the leakage source probability distribution map according to the updated posterior probabilities of the sub-areas, the method further includes: Local refinement search is used to improve the grid resolution in each sub-region marked by the binary thermal imaging mask. The posterior probability of each sub-region is optimized according to the gas concentration gradient, and the sequential Monte Carlo method is used to update the leakage source probability distribution map at preset intervals.

6. The method for dynamic path planning of a UAV cluster based on a Gaussian plume model according to claim 1, characterized in that: The method of planning the search path of the drone cluster based on the concentration gradient value and the pheromone field strength includes: Determine the probability of the drone moving: ； Where, Indicates the drone is moving from its current position Move to adjacent position probability; Represents concentration gradient, reaction position The rate of change of gas concentration; Indicates location The pheromone strength, Indicates that there is adjacent locations.

7. The method for dynamic path planning of a UAV cluster based on a Gaussian plume model according to claim 1, characterized in that: The gridding of the leakage source probability distribution map to determine the search tasks and the allocation of the search tasks according to the auction algorithm and the real-time status of the drone include: Divide the leakage source probability distribution map into several sub-areas by gridding, and mark the leakage probability value for each sub-area, where each sub-area corresponds to a search task; Obtain real-time drone status data, including the remaining battery life and the response time required to scan a sub-area; Determine the bidding function of the auction algorithm based on the real-time status data of the drone: ； Where, For drones is the bid price of the sub-region, For drones The remaining power, is the response time of scanning sub-areas, is the leakage probability of the sub-area. The higher the leakage probability, the higher the bid price. Search tasks are assigned according to the bidding function, and the task assignment results are optimized with the goal of minimizing total time consumption and balancing energy consumption.

8. A UAV swarm dynamic path planning system based on Gaussian plume model, characterized by: The system comprises: A data acquisition unit, used to collect gas leakage concentration, wind speed data, and thermal imaging data from chemical plants through a swarm of drones; A probability distribution determination unit is used to dynamically modify the diffusion coefficient of the Gaussian plume model based on wind speed data, and to generate a leakage source probability distribution map based on the leakage gas concentration, thermal imaging data, and the modified Gaussian plume model; The search path planning unit is used to determine the concentration gradient value and pheromone field strength at each location of the chemical plant based on the leakage source probability distribution map, and plan the search path of the drone cluster based on the concentration gradient value and pheromone field strength; A search task allocation unit is used to grid the leakage source probability distribution map to determine the search tasks, allocate the search tasks according to the auction algorithm and the real-time status of the UAV, and search and locate the leakage source according to the search path and the allocated search tasks; The leakage point marking unit is used to mark the leakage point according to the leakage source positioning results and transmit the information back to the command center of the drone cluster in real time.

9. An electronic device, characterized in that: include: A processor and a memory, the memory being used to store computer program code, the computer program code including computer instructions, and when the processor executes the computer instructions, the electronic device executes the method for dynamic path planning of a drone cluster based on a Gaussian plume model as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor of an electronic device, the processor executes the method for dynamic path planning of a drone cluster based on a Gaussian plume model as described in any one of claims 1 to 7.

Images