Multi-robot smell source positioning method based on fuzzy reasoning parameter adaptive control

Abstract

The invention provides a multi-robot smell source positioning method based on fuzzy reasoning parameter adaptive control, and aims to realize parameter adaptive adjustment in a smell source positioning task of multiple robots based on an improved particle swarm optimization algorithm of fuzzy reasoning and improve the search efficiency, the success rate and the environmental adaptability. Comprising the steps of establishing a smell source positioning simulation environment model, designing an improved particle swarm optimization algorithm, designing a fuzzy controller and positioning a smell source. According to the method, parameter adaptive control is realized through a fuzzy reasoning mechanism, and the intelligence, the stability and the robustness of the source positioning system in a complex plume environment are effectively improved in combination with the improved particle swarm optimization algorithm with enhanced guide items. The method is simple in algorithm structure and excellent in performance, and has wide simulation application potential and engineering popularization value.

Description

Technical Field

[0001] The present invention belongs to the technical field of odor source positioning, and in particular relates to a multi-robot odor source positioning method based on fuzzy inference parameter adaptive control. Background Art

[0002] Odor source localization is a critical issue in environmental monitoring, disaster response, and pollution tracing. Its core goal is to accurately identify pollution or leak sources in complex diffusion environments based on gas concentration distribution characteristics. With the advancement of environmental simulation and intelligent algorithm technologies, a growing number of studies are using simulation modeling to simulate odor diffusion, leveraging multiple "virtual robots" or "particle agents" to collaboratively search within the simulation space to rapidly locate odor sources. In this context, swarm intelligence optimization methods, particularly particle swarm optimization (PSO), have become a research hotspot for source localization due to their simple structure, strong collaboration, and excellent parallelism. These methods use odor concentration as a fitness function and iteratively update particle positions to approximate the source, achieving promising initial results.

[0003] At present, existing PSO methods still have many shortcomings in complex environments. First, traditional PSO uses fixed parameters and lacks the ability to adaptively adjust to different diffusion stages and airflow conditions. It is easy to fall into local optimality in situations such as turbulent interference and plume breakup, affecting search efficiency. Second, although some studies have attempted to introduce wind field information to enhance search guidance, most of them are static corrections or empirical controls, and fail to fully integrate the combined effects of concentration changes and wind direction dynamics. In addition, most current parameter adjustment strategies still rely on manual experience and are difficult to generalize to different environments, which seriously restricts the application effect of swarm intelligence methods in real or simulated environments.

[0004] To address the above problems, it is necessary to invent an intelligent optimization method that can automatically adjust parameters according to search behavior and adapt to different complex environments, so as to improve the robustness, accuracy and efficiency of the source positioning system in complex scenarios. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a multi-robot odor source localization method based on fuzzy reasoning parameter adaptive control, aiming to realize parameter adaptive adjustment in the odor source localization task performed by multiple robots based on an improved particle swarm optimization algorithm based on fuzzy reasoning, thereby improving search efficiency, success rate and environmental adaptability.

[0006] The present invention provides a multi-robot odor source localization method based on fuzzy inference parameter adaptive control, comprising the following steps:

[0007] (1) Construction of odor source positioning simulation environment model

[0008] The simulation environment model uses a two-dimensional coordinate plane to represent the search area, and periodically releases simulated odor particles from a fixed source to form a smoke plume trajectory. The environment model also simulates the odor sensor, anemometer, and positioning module. The airflow field in the environment model is generated by a model consisting of average wind speed and Gaussian perturbations.

[0009] Furthermore, in the airflow field, the wind speed vector Expressed as:

[0010]

[0011] represents the coordinates of any point in space, Indicates the current time; is the constant mean wind speed vector; is Gaussian white noise disturbance, where is the noise variance. Turbulence intensity Defined as:

[0012]

[0013] is the standard deviation of wind speed component, ,in and is the standard deviation of airflow velocity in the x and y directions; is the average wind speed, ,in and yes and The average value of .

[0014] In the simulation environment model, the odor source is located at a fixed position in the two-dimensional plane and releases a fixed number of plume particle packets into the environment every second. After release, each particle packet translates with the local wind speed field, simulating the process of odor clusters being blown by wind and drifting and diffusing over time in a real environment. The formula is as follows:

[0015]

[0016] in, is the position vector of the ith package, U is the dynamic airflow velocity, is the molecular diffusion noise;

[0017] The robot uses the odor sensor to read the particle packet contact situation near the current position to obtain the concentration value. If the concentration exceeds the perception threshold, it is considered to have detected the odor. The concentration of any point (x, y) in space at time t is By summing up all packages:

[0018]

[0019] in, is the number of active packages at time t, is the Gaussian kernel function, and the formula is as follows:

[0020]

[0021] in, Represents the Euclidean distance from the target point to the current position of the filamentous package in space; is the time that has passed since the package was released; , is the diffusion radius, and D is the molecular diffusion coefficient.

[0022] Turbulence causes the package to break, which is controlled by the life decay function. After disappearing, the simulated plume breaks; is the life decay function, the formula is:

[0023]

[0024] in, is the initial lifespan, is the turbulence intensity coefficient, is the velocity gradient mode; the diffusion coefficient in turbulent environment is corrected to the following formula:

[0025]

[0026] in, is the turbulent diffusion coefficient.

[0027] (2) Improved particle swarm optimization algorithm design

[0028] The Particle Swarm Optimization (PSO) algorithm uses the fitness value to evaluate whether the robot is in an appropriate position. The fitness value at time t is given by Said that this is In the standard PSO algorithm, the position and velocity of each robot are updated to improve the fitness value, which is expressed as:

[0029]

[0030] in and are the position vector and velocity vector of the i-th robot at time t respectively; is the time step, 、 、 is a dimensionless parameter, is the inertia weight, 、 They are individual learning factor and group learning factor respectively; and are the optimal local position of the i-th robot during the period from 0 to t and the optimal global position of the robot swarm, respectively. In the improved particle swarm algorithm, each robot updates its position and velocity by adding a headwind term so as to move towards areas and sources with higher concentrations. The velocity of the i-th robot is updated as follows:

[0031]

[0032] in, and 、 Similarly, it is a dimensionless factor that reflects the effect of airflow on the robot's speed, that is, the windward coefficient; 、 、 is a random number uniformly distributed in the range [0,1]; is the headwind speed of the i-th robot at the t-th moment, expressed as:

[0033]

[0034] in, is the maximum moving speed of the robot (m / s); is the airflow velocity of the i-th robot at time t (m / s); is the threshold of the airflow sensor; is a random direction vector uniformly distributed in the range [-1,1], which is added to the headwind speed when the airflow speed is lower than the sensor threshold;

[0035] (3) Fuzzy controller design

[0036] The fuzzy reasoning process is divided into five main steps: input collection, fuzzification, fuzzy rule matching, fuzzy reasoning, and defuzzification;

[0037] The fuzzy controller sets three input variables and four output variables. The input variables in the input acquisition include turbulence intensity, current detection concentration and concentration change rate;

[0038] The fuzzification method uses triangular and trapezoidal functions to fuzzily divide continuous input quantities and map them into fuzzy language values ​​with a membership degree between 0 and 1. The turbulence intensity level is set to high, medium, and low, the current concentration level is set to high and low, and the concentration change rate level is set to increase and decrease.

[0039] Fuzzy rules are summarized through experimental statistics and behavioral analysis;

[0040] Fuzzy output variables include inertia weights , individual learning factor , group learning factor and headwind coefficient The fuzzy levels are small, medium, and large. The real values ​​of the four parameters are obtained through the defuzzification method of the Mamdani inference mechanism and the center of mass method, which are used in the next velocity update of the particle.

[0041] (IV) Odor source localization process

[0042] The odor source localization task is divided into three stages:

[0043] (1) Source discovery phase: Using a zigzag search strategy, all particles initially move in random directions and at random speeds in space. They determine whether they have entered the plume area by detecting whether the odor concentration exceeds a threshold. When any particle detects a concentration greater than the threshold, all particles enter the next phase.

[0044] (2) Plume tracking stage: start the improved PSO algorithm and call the fuzzy controller at the same time to automatically adjust the parameters according to the particle behavior state of each iteration 、 、 , guiding the particles to track the plume along the concentration gradient and upwind direction, quickly approaching the source; once the plume disappears during the crossing process, the robot will move randomly to find the plume again;

[0045] (3) Source confirmation stage: Using the maximum concentration method, when the global optimal point of the particle swarm does not change in a number of consecutive iterations, it is determined that the source point has been located and the search is terminated.

[0046] By comparing the performance of various methods in terms of search success rate, average positioning error, convergence time and other indicators, it is verified that the method of the present invention has a significant performance improvement effect in the simulation environment, especially in scenarios such as strong turbulence, weak source signals, and plume distortion, it shows higher robustness and intelligence.

[0047] Beneficial effects of the present invention:

[0048] 1. Adaptive adjustment of particle swarm optimization parameters to improve environmental adaptability and search efficiency. This invention implements online adaptive adjustment of key parameters of the particle swarm optimization algorithm by constructing a fuzzy inference control system based on search behavior variables such as particle dispersion, local concentration gradient, and turbulence intensity. This mechanism enables the search algorithm to dynamically adjust its strategy based on the current search state, maintaining efficient search performance under varying diffusion conditions, plume morphologies, and signal strengths, significantly outperforming traditional optimization methods that use fixed parameters.

[0049] Second, introducing a wind direction guidance term enhances particle escape capabilities and overcomes local optimal traps. This invention introduces an upwind guidance term based on wind direction information into the PSO algorithm, enabling particles to perform targeted plume tracing in an upwind direction when concentration information is insufficient or the search is stagnant. This improvement effectively overcomes the problem of traditional PSO being prone to falling into local optimality and chaotic search paths in complex wind fields. In particular, in simulation scenarios with strong turbulence and significant plume breakage, the upwind term can significantly improve positioning success rate and escape efficiency.

[0050] Third, a simulation environment based on a particle release model was established to realistically reproduce the intermittent characteristics of plumes. This paper employs an odor source modeling approach based on the filament particle method, combined with an airflow model of average wind speed and Gaussian perturbations, to realistically simulate the intermittent and irregular diffusion of odor plumes in two-dimensional space. This simulation platform supports modeling of high-frequency odor release, wind speed variations, and sensor noise, comprehensively accounting for the uncertainty of the perceived signal and environmental perturbations, providing a reliable foundation for evaluating the robustness and universality of source localization strategies.

[0051] Fourth, construct a phased source localization process to improve the systematicness and stability of search decisions. This invention divides the entire source localization process into three stages: the source discovery stage uses a Z-shaped search strategy to expand coverage; the plume tracking stage uses an improved PSO algorithm with fuzzy inference control to accelerate convergence; and the source confirmation stage uses a maximum concentration determination method to confirm the source point location. This process combines the advantages of exploration and convergence, making the search strategy more hierarchical and the switching more reasonable, thereby improving the stability and controllability of the overall search system.

[0052] Fifth, it provides a highly versatile and easily integrated algorithmic framework suitable for a variety of simulation applications. This invention is independent of actual robotic platforms or hardware devices; all algorithms run in a simulation environment, making it suitable for a variety of application scenarios, including environmental modeling, pollution source tracking, and virtual robot behavior simulation. The proposed method exhibits excellent portability and scalability, seamlessly integrating with CFD simulation platforms, sensor noise models, and swarm intelligence testing frameworks. It provides a theoretical foundation and technical support for intelligent source identification tasks in multi-source, multi-obstacle, and high-disturbance scenarios.

[0053] In summary, this method achieves adaptive parameter control through a fuzzy inference mechanism, combined with an improved particle swarm optimization algorithm enhanced with guidance terms, effectively improving the intelligence, stability, and robustness of the source localization system in complex plume environments. This method, with its simple algorithmic structure and excellent performance, has broad potential for simulation applications and engineering promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of the simulation environment model of the present invention;

[0055] Figure 2 Schematic diagram of the fuzzy controller framework of the present invention;

[0056] Figure 3 A schematic diagram of the Z-shaped search strategy of the present invention; is the extreme value of the search area boundary; To search for the initial position;

[0057] Figure 4 Schematic diagram of the overall traceability process of the present invention. DETAILED DESCRIPTION

[0058] The present invention provides a multi-robot odor source localization method based on fuzzy inference parameter adaptive control, comprising the following steps:

[0059] (1) Construction of odor source positioning simulation environment model

[0060] The present invention uses a two-dimensional coordinate plane to represent the search area of ​​the environmental model. Simulated odor particles are periodically released from a fixed source to form a plume trajectory. The environmental model also simulates an odor sensor, anemometer, and positioning module. The airflow field in the environmental model is generated using a model composed of average wind speed and Gaussian perturbations. Its turbulence intensity can be controlled to simulate the actual conditions of plume distortion and intermittent diffusion in complex environments.

[0061] Furthermore, in the airflow field, the wind speed vector Expressed as:

[0062]

[0063] represents the coordinates of any point in space, Indicates the current time; is the constant mean wind speed vector, (1, 0) m / s for laminar flow and (2, 0) m / s for turbulent flow; is Gaussian white noise disturbance, where is the noise variance, laminar flow =0, turbulent flow = 20. Turbulence intensity Defined as:

[0064]

[0065] is the standard deviation of wind speed component, ,in and is the standard deviation of airflow velocity in the x and y directions; is the average wind speed, ,in and yes and The average value of .

[0066] In the simulation environment model, the odor source is located at a fixed position on a two-dimensional plane and releases a fixed number of plume particle packets into the environment every second. After release, each particle packet translates with the local wind speed field, simulating the process of odor clusters being blown by wind and drifting and diffusing over time in a real environment.

[0067] In the simulation environment of this embodiment, the odor source is located at the position (20, 0) on the two-dimensional plane. N = 10 filamentous plume packets are released into the environment per second. Each packet carries a fixed odor molecular weight q. After release, each particle packet moves horizontally with the local wind speed field, simulating the process of odor clusters being blown by wind and drifting and diffusing over time in a real environment. The formula is as follows:

[0068]

[0069] in is the position vector of the ith package, U is the dynamic airflow velocity, is the molecular diffusion noise.

[0070] The discontinuous plume trajectory created by this particle release model is highly patchy and intermittent, effectively simulating odor signals under turbulent conditions. The robot uses an odor sensor to read the particle packet contact information near its current location and obtains a concentration value. If the concentration exceeds the perception threshold, the odor is considered detected. The odor concentration and wind speed data collected by the odor sensor and anemometer serve as algorithm inputs to drive the fuzzy controller and source search strategy, constructing a complete odor source location simulation environment model.

[0071] The concentration of any point (x,y) in space at time t By summing up all packages:

[0072]

[0073] in, is the number of active packages at time t, is the Gaussian kernel function, and the formula is as follows:

[0074]

[0075] in, Represents the Euclidean distance from the target point to the current position of the filamentous package in space; is the time that has passed since the package was released; , is the diffusion radius, and D is the molecular diffusion coefficient.

[0076] Turbulence causes the package to break, which is controlled by the life decay function. After disappearing, the simulated plume breaks; is the life decay function, the formula is:

[0077]

[0078] in, is the initial lifespan, is the turbulence intensity coefficient, is the velocity gradient mode; the diffusion coefficient in turbulent environment is corrected to the following formula:

[0079]

[0080] in, is the turbulent diffusion coefficient;

[0081] The basic environment model of this example is as follows Figure 1 As shown, the size is 100*100 , there are no reflective boundaries, the wind direction is +x, the odor source is fixed at (20, 0) m, and N = 10 filamentous plume packets are released into the environment per second. Each robot is considered a particle, and its own impact on source tracing is not considered. The Gaussian noise of the concentration sensor, anemometer, and positioning sensor is set to 0.05 ppm, 0.1 m / s, and 0.1 m, respectively. The sampling frequency is set to 100 Hz, and the plume threshold is found to be 0.1 ppm; When the velocity is (1,0) m / s, the flow environment is laminar; when the velocity is (2,0) m / s, the flow environment is turbulent. The molecular diffusion coefficient is set to 0.01, and the turbulent diffusion coefficient and attenuation coefficient are set to 0.5 and 0.2, respectively.

[0082] (2) Improved particle swarm optimization algorithm design

[0083] The Particle Swarm Optimization (PSO) algorithm uses the fitness value to evaluate whether the robot is in an appropriate position. The fitness value at time t is given by Said that this is In the standard PSO algorithm, the position and velocity of each robot are updated to improve the fitness value, which is expressed as:

[0084]

[0085] in and are the position vector and velocity vector of the i-th robot at time t respectively; is the time step, 、 、 is a dimensionless parameter, is the inertia weight, 、 are the individual learning factor and the group learning factor, respectively. In this embodiment, they are generally set to 1.0, 2.0, and 2.0; and are the optimal local position of the i-th robot during the period from 0 to t and the optimal global position of the robot swarm, respectively. In the improved particle swarm algorithm, each robot updates its position and velocity by adding a headwind term so as to move towards areas with higher concentrations and sources; the velocity of the i-th robot is updated as:

[0086]

[0087] in, and 、 Similarly, it is a dimensionless factor reflecting the effect of airflow on the robot's speed, that is, the windward coefficient, which is set to 2.0 in this embodiment; 、 、 is a random number uniformly distributed in the range [0,1]; is the headwind speed of the i-th robot at the t-th moment, expressed as:

[0088]

[0089] in, is the maximum moving speed of the robot (m / s); is the airflow velocity of the i-th robot at time t (m / s); is the threshold of the airflow sensor; is a random direction vector uniformly distributed in the range [-1,1], which is added to the headwind speed when the airflow speed is below the sensor threshold.

[0090] (3) Fuzzy controller design

[0091] In order to improve the algorithm's ability to respond to environmental conditions, the present invention introduces a fuzzy inference control system to adjust the particle swarm parameters online. The fuzzy inference process is divided into five main steps: input acquisition, fuzzification, fuzzy rule matching, fuzzy inference, and defuzzification. The fuzzy controller block diagram is shown in Figure 1. Figure 2 shown.

[0092] The fuzzy controller is set with three input variables and four output variables. The input variables in the input acquisition include turbulence intensity (determining the stability and reaction speed of the overall search strategy), current detection concentration (estimating whether the particles are close to the odor source and determining the convergence or diffusion tendency), and concentration change rate (determining the local search trend and used for fine-grained parameter adjustment dynamics);

[0093] The fuzzification method uses triangular and trapezoidal functions to fuzzily divide continuous input quantities and map them into fuzzy language values ​​with a membership degree between 0 and 1. The turbulence intensity level is set to high, medium, and low, the current concentration level is set to high and low, and the concentration change rate level is set to increase and decrease.

[0094] The fuzzy rules are summarized through experimental statistics and behavioral analysis, with a total of 3*2*2=12 fuzzy rules. The experimental environment includes three turbulence intensities and their corresponding two current concentrations, a total of six;

[0095] The specific experimental process is as follows:

[0096] First, all combinations of the four parameters are traversed. Each experiment uses the improved PSO algorithm with different sets of parameters to conduct 10 simulation experiments. The average search time, success rate and positioning accuracy of each set of experiments and the concentration changes of each round of iteration are recorded. Finally, the best parameter combination is obtained in each experimental environment, and these combinations are reflected in the fuzzy output. The fuzzy output variables include inertia weight , individual learning factor , group learning factor and headwind coefficient The fuzziness levels are small, medium, and large. The real values ​​of the four parameters are obtained through the Mamdani inference mechanism and the center of mass method defuzzification method, which are used in the next particle velocity update.

[0097] The ultimate implementation process uses the three input variables detected during source tracing to determine the optimal control parameters. For example, if low turbulence intensity, high current concentration, and increasing concentration change rate are detected, the inertia weight is reduced, the learning factor is increased, and the headwind coefficient is reduced. This indicates that the current position is close to the source and the inertia should be reduced to achieve appropriate convergence.

[0098] A large number of simulations were conducted using the improved PSO algorithm under different simulation environment model settings. The corresponding optimal parameter combinations were recorded and a fuzzy control rule library was constructed. The inference mechanism used the Mamdani model. The output parameters were solved through membership function and defuzzified before being input into the improved PSO algorithm to achieve adaptive adjustment.

[0099] (IV) Odor source localization process

[0100] The present invention divides the odor source localization task into three stages:

[0101] (1) Source discovery phase: Using a zigzag search strategy, all particles initially move in random directions and at random speeds in space. They determine whether they have entered the plume area by detecting whether the odor concentration exceeds a threshold. When any particle detects a concentration greater than the threshold, all particles enter the next phase.

[0102] (2) Plume tracking stage: start the improved PSO algorithm and call the fuzzy controller at the same time to automatically adjust the parameters according to the particle behavior state of each iteration 、 、 , guiding the particles to track the plume along the concentration gradient and upwind direction, quickly approaching the source; once the plume disappears during the crossing process, the robot will move randomly to find the plume again;

[0103] (3) Source confirmation stage: Using the maximum concentration method, when the global optimal point of the particle swarm does not change in a number of consecutive iterations, it is determined that the source point has been located and the search is terminated.

[0104] This process can cope with multi-source, multi-ventilation field, and multi-obstacle simulation scenarios, and has good environmental adaptability and versatility.

[0105] (V) Experimental verification

[0106] In order to verify the effectiveness and advancement of the proposed method, the present invention designs multiple sets of source localization simulation experiments. The following are respectively set:

[0107] Different types of source terms (periodic and decaying);

[0108] Different wind field structures (mixing ventilation, displacement ventilation);

[0109] Different turbulence intensities and spatial obstacle layouts;

[0110] In each set of experiments, the following algorithms are compared:

[0111] Standard PSO (fixed parameters);

[0112] Improved PSO (adding upwind terms but with fixed parameters);

[0113] Wind direction utilization algorithm (other multi-robot source localization methods);

[0114] The method of the present invention (improved PSO + fuzzy reasoning adaptation);

[0115] Experimental results were obtained by comparing various methods in terms of search success rate, average positioning error, convergence time, and other indicators. The final results show that the proposed method achieves high positioning success rate and high positioning accuracy in both environments. In a laminar flow environment, the time required for tracing the source is approximately 40% shorter than that of the standard PSO method, approximately 36% shorter than that of the improved PSO algorithm, and approximately 60% shorter than that of the wind direction-based algorithm. These results confirm that the proposed method has a significant performance improvement in a simulation environment.

[0116] The multi-robot odor source localization method of the present invention can be widely applied in many fields. The following are several specific application examples:

[0117] 1. Simulation and positioning of toxic gas leaks in chemical parks

[0118] This invention can be applied to source-tracing simulations in the event of a toxic gas leak in a chemical plant or petrochemical industrial park. By constructing a two-dimensional or three-dimensional spatial model of the park, the diffusion of gas in a wind field is simulated, and the proposed fuzzy-reasoning enhanced particle swarm optimization algorithm is used to locate the source under varying turbulence intensities. Compared to traditional fixed-parameter methods, this method exhibits greater search robustness and rapid response capabilities in complex environments such as wind speed fluctuations and plume breakups, making it suitable for emergency response drill systems and virtual accident assessment platforms.

[0119] 2. Urban Air Pollution Source Tracing Simulation

[0120] In urban environmental monitoring, this method can be used to build air quality simulation systems. By simulating pollution source emissions and wind diffusion, it can intelligently locate unknown pollution sources. By integrating meteorological data (such as wind speed, direction, and temperature) and pollutant simulation fields, this method can efficiently track and locate pollution events such as industrial waste gas and traffic emissions, providing support for environmental law enforcement, air quality forecasting, and pollution control measures.

[0121] 3. Location of Abnormal Sources of Harmful Gases in Underground Transportation Systems

[0122] This system can simulate odor source location in ventilation-restricted spaces such as subways, tunnels, and underground garages, and is suitable for simulating the source identification of toxic and flammable gases or oxygen-deficient areas. The system generates complex airflow paths using wind field disturbance and diffusion models, and then searches for pollution sources using an improved particle swarm algorithm. In simulated environments with high turbulence and strong sensory interference, this system can quickly locate the source, improving accident investigation efficiency and making it suitable for underground space safety simulation training systems.

[0123] 4. Pollution source inversion simulation training under disaster scenarios

[0124] This invention can be used to build a pollution source location simulation system for disaster scenarios such as chemical leaks and biological contamination after earthquakes. By setting up complex environmental conditions such as multiple obstacles, intermittent signals, and strong disturbances, the intelligent agent's ability to locate sources in uncertain scenarios can be tested. This system can be used for tasks such as emergency response strategy training and unmanned search strategy algorithm verification, demonstrating its value in virtual simulation training and strategy comparison research.

[0125] These application examples fully demonstrate the high adaptability and robustness of the present invention in complex, dynamic, and multi-disturbance environments, and demonstrate its wide applicability and engineering promotion value in multiple fields such as virtual simulation, pollution source tracing, and intelligent search and control.

Claims

1. A multi-robot odor source localization method based on fuzzy inference parameter adaptive control, characterized by: The following steps are involved: Step 1: Build a simulation environment model for odor source positioning The simulation environment model uses a two-dimensional coordinate plane to represent the search area, and periodically releases simulated odor particles from a fixed source point to form a smoke plume trajectory. The environment model also simulates the odor sensor, anemometer, and positioning module. The airflow field in the environment model is generated using a model consisting of average wind speed and Gaussian perturbations. Step (2): Improve the design of particle swarm optimization algorithm The particle swarm optimization algorithm (PSO) uses the fitness value to evaluate whether the robot is in an appropriate position; the i-th robot The fitness value at time t is given by Said that this is The concentration of the measurement; in the standard PSO algorithm, the position and velocity of each robot are updated to improve the fitness value, which is expressed as: ， in and are the position vector and velocity vector of the i-th robot at time t respectively; is the time step, 、 、 is a dimensionless parameter, is the inertia weight, 、 They are individual learning factor and group learning factor respectively; and are the optimal local position of the i-th robot during the period from 0 to t and the optimal global position of the robot swarm, respectively. In the improved particle swarm algorithm, each robot updates its position and velocity by adding a headwind term so as to move towards areas and sources with higher concentrations. The velocity of the i-th robot is updated as follows: ， in, and 、 Similarly, it is a dimensionless factor that reflects the effect of airflow on the robot's speed, namely the windward coefficient; 、 、 is a random number uniformly distributed in the range [0,1]; is the headwind speed of the i-th robot at the t-th moment, expressed as: ， in, is the maximum moving speed of the robot (m / s); is the airflow velocity of the i-th robot at time t (m / s); is the threshold of the airflow sensor; is a random direction vector uniformly distributed in the range [-1,1], which is added to the headwind speed when the airflow speed is lower than the sensor threshold; Step 3: Fuzzy controller design The fuzzy reasoning process is divided into five main steps: input collection, fuzzification, fuzzy rule matching, fuzzy reasoning, and defuzzification; The fuzzy controller sets three input variables and four output variables. The input variables in the input acquisition include turbulence intensity, current detection concentration and concentration change rate; The fuzzification method uses triangular and trapezoidal functions to fuzzily divide continuous input quantities and map them into fuzzy language values ​​with a membership degree between 0 and 1. The turbulence intensity level is set to high, medium, and low, the current concentration level is set to high and low, and the concentration change rate level is set to increase and decrease. Fuzzy rules are summarized through experimental statistics and behavioral analysis; Fuzzy output variables include inertia weights , individual learning factor , group learning factor and headwind coefficient The fuzziness levels are small, medium, and large. The real values ​​of the four parameters are obtained through the Mamdani inference mechanism and the center of mass method defuzzification method, which are used in the next particle velocity update. Step 4: Odor source location process The odor source localization task is divided into three stages: (1) Source discovery phase: Using a zigzag search strategy, all particles initially move in random directions and at random speeds in space. They determine whether they have entered the plume area by detecting whether the odor concentration exceeds a threshold. When any particle detects a concentration greater than the threshold, all particles enter the next phase. (2) Plume tracking stage: start the improved PSO algorithm and call the fuzzy controller at the same time to automatically adjust the parameters according to the particle behavior state of each iteration 、 、 , guiding the particles to track the plume along the concentration gradient and upwind direction, quickly approaching the source; once the plume disappears during the crossing process, the robot will move randomly to find the plume again; (3) Source confirmation stage: Using the maximum concentration method, when the global optimal point of the particle swarm does not change in a number of consecutive iterations, it is determined that the source point has been located and the search is terminated.

2. The multi-robot odor source localization method based on fuzzy inference parameter adaptive control according to claim 1 is characterized by: In the airflow field of step (1), the wind speed vector Expressed as: ， represents the coordinates of any point in space, Indicates the current time; is the constant mean wind speed vector; is Gaussian white noise disturbance, where is the noise variance; turbulence intensity Defined as: ， is the standard deviation of wind speed component, ,in and is the standard deviation of the airflow velocity in the x and y directions; is the average wind speed, ,in and yes and The average value of In the simulation environment model, the odor source is located at a fixed position in the two-dimensional plane and releases a fixed number of plume particle packets into the environment every second. After release, each particle packet translates with the local wind speed field, simulating the process of odor clusters being blown by wind and drifting and diffusing over time in a real environment. The formula is as follows: ， in, is the position vector of the ith package, U is the dynamic airflow velocity, is the molecular diffusion noise.

3. The multi-robot odor source localization method based on fuzzy inference parameter adaptive control according to claim 1 or 2, characterized in that: In step (1), the robot uses the odor sensor to read the particle packet contact situation near the current position to obtain the concentration value. If it exceeds the concentration perception threshold, it is considered that the odor is detected. The concentration of any point (x, y) in space at time t is By summing up all packages: ， in, is the number of active packages at time t, is the Gaussian kernel function, and the formula is as follows: ， in, Represents the Euclidean distance from the target point to the current position of the filamentous package in space; is the time that has passed since the package was released; , is the diffusion radius, D is the molecular diffusion coefficient; Turbulence causes the package to break, which is controlled by the life decay function. After disappearing, the simulated plume breaks; is the life decay function, the formula is: ， in, is the initial lifespan, is the turbulence intensity coefficient, is the velocity gradient mode; the diffusion coefficient in turbulent environment is corrected to the following formula: ， in, is the turbulent diffusion coefficient.

Images