Hybrid Strategy Improved Grey Wolf Optimization Algorithm for Plume Tracking and Localization Method in Indoor Weak Wind Environment

Reliable tracking of the plume by the robot is the key to achieving plume source localization. To address the problem of low success rate and long search time of robot source location due to the unavailability of reliable information on gas diffusion flow direction and flow velocity in an indoor weak wind environment, a hybrid strategy is proposed to improve the grey wolf optimization algorithm for robot plume tracking and location. The plume is modeled using Computational Fluid Dynamics (CFD) in a two-dimensional indoor weak wind environment, and the plume concentration value is used as the individual adaptation degree of the algorithm. Without carrying plume velocity and flow direction sensors, the source-finding robot simulates the grey wolf population’s social mechanism and hunting behavior to update its position. The improved Grey Wolf Optimization algorithm is compared with the traditional Grey Wolf Optimization (GWO), Particle Swarm Algorithm (PSO), and Genetic Algorithm (GA) in simulation experiments. The simulation experiments show that the average number of iterations of the improved GWO is 7, 108, and 118 times shorter than the four source finding algorithms of GWO, PSO, and GA. The average planning path is reduced by 0.91 meters, 2.35meters, and 2.90 meters. 3s, 10.3s, and 9.3s reduce the average running time. The average positioning success rate is improved by 30%, 32%, and 40%. The applicability and Stability of the improved GWO algorithm in solving the plume tracking and localization problem in an indoor weak wind environment are verified.

to numerous accidents, such as fires and explosions. Major The associate editor coordinating the review of this manuscript and approving it for publication was Yingxiang Liu . on society and the environment [1], [2]. To accurately locate 28 the source of contamination within a short time, effectively 29 control the scope of contamination, make timely warnings 30 and evacuate personnel, and reduce economic and property 31 losses to the greatest extent possible [3]. Research on how to 32 locate the source of flammable and explosive toxic gas leaks 33 has become an urgent problem in the field of public safety. 34 A plume [4] is a material plume created by the spread of a 35 spill through a medium. Rapid identification of indoor plume 36 sources, based primarily on plume concentration distribution, 37 is a requirement and fundamental to effective control and 38 mitigation strategies [5], [6]. The identification of the release 39 intensity of time-varying pollutants with a known indoor 40 source site was studied in the literature [7], [8], [9]. [10]   The indoor plume concentration distribution is mainly gov-66 erned by the indoor flow field, the location of the plume 67 source release, and the intensity of the plume source release.

68
In the case of an indoor steady-state flow field where the 69 area of the plume source is known, the pollutant concentra-70 tion distribution is governed by the intensity of the plume 71 source release [8]. In practice, the location of the head 72 may be unknown, considering that wind direction and speed 73 information is not significant in a weak wind field or free 74 diffusion environment. This paper uses CFD to establish a 75 two-dimensional indoor plume concentration distribution in 76 a soft wind environment. By analyzing the implementation 77 mechanism of GWO, a hybrid strategy is proposed to improve 78 the grey wolf optimization algorithm for robot plume tracking 79 and localization. The source-finding robot uses the gas con-     where u is the x-axis velocity component, m/s; v is the y-axis 118 velocity component, m/s; w is the z-axis velocity component, 119 m/s; p is the gas density, kg/m 3 ; u i is the i direction velocity 120 component, m/s; u i is the j direction velocity component, m/s; 121 P is the pressure, pa; R is the gas constant of a gas, J(kg·k); u is 122 the kinetic viscosity, T is the thermodynamic temperature, k; 123 α is the thermal diffusion coefficient, m 2 /s; k is the turbulent 124 kinetic energy, m 2 /s 2 ; u eff is the effective kinetic viscosity, 125 kg/(m·s); ξ is the dissipation rate, m 2 /s 3 ; G k is the term for the 126 generation of rough kinetic energy k due to the mean velocity 127 gradient, kg/(m·s 3 ); αε is the Planter number corresponding 128 to the dissipation rate; u t is the turbulent kinetic viscosity, 129 kg/(m·s); I are the tensor symbols, which take values in the 130 range 1, 2.  chamber according to the actual calculation requirements.

135
One H2S leaking device is inserted as a simulated leakage 136 source in this paper's simulation. Figure 1 shows the model  expressions (8) to (12).
where: t denotes the current number of iterations; D denotes 161 the distance between the gray wolf and the prey; X p (t) is the 162 position of the game in the t generation; X i (t) is the position 163 vector of a single gray wolf in the first generation; A and C 164 is the random constant coefficient, which is calculated as: r 1 and r 2 is the random number between; the value decreases 168 from 2 to 0 as the number of iterations increases, the expres-169 sion is: where t is the current number of iterations; t max is the maxi-172 mum number of iterations. .
where: X α , X β and X δ are the α wolf, β wolf and δ wolf posi- This paper addresses the shortcomings of the grey wolf 216 algorithm, which is prone to local optimal and slow 217 VOLUME 10, 2022  (20): The rules for meta-cell evolution are:

248
(1) Based on the extended Moore neighborhood model in 249 figure 5, the central cell of the shaded area is selected, 250 and the optimum plume concentration (fitness) for all 251 regions is calculated as the initial value, where s iq is the judged value of the plume concentration.

254
(2) For a single region M i take any of the cells 255        To avoid the influence of randomness on the simulation 296 experiments, the indoor plume concentration fields at T = 297 300s and T = 420s were taken as the simulation cloud envi-298 ronment for the four source-finding algorithms in the plume 299 CFD simulation model. 50 plume tracking and positioning 300 simulation experiments were carried out in Simulation soft-301 ware using the source-finding robot. The initial position of 302 it is considered a failure. In the simulation experiments, the 319 source-seeking robot is approximated as a point, which is the 320 ''ideal robot.'' The source-seeking robot can know its own 321 coordinates and accurately follow the path planned by the 322 source-seeking algorithm to track and locate the plume.

323
As seen in figure 7 the GWO algorithm successfully 324 located the plume source after 27 and 23 position updates 325 in the plume turbulence environment at T = 300s and T = 326 420s, respectively, without relying on the plume velocity and 327 flow direction sensors. As seen in figure 7, the concentration 328 changes over a more extended period are small, so the original 329 grey wolf algorithm still suffers from slow convergence and 330 falls into a local optimum, leading to a low success rate in the 331 final plume tracking and localization. the GA algorithm and preserves the optimal solution. Still, 339 its convergence is random, the search tends to stall, and the 340 algorithm tends to fall into a local optimum, resulting in a low 341 success rate of the PSO algorithm in source finding.

342
As seen in figure 9, the plume source was successfully The improved grey wolf algorithm makes the particles ini-358 tially more homogeneous due to a good set of points for their 359 initialization. At the same time, the target locations across 360 the plume range become easy to determine using cellular 361 automation, avoiding premature convergence of the wolf pack 362 and trapping the algorithm in a local optimum.

363
In summary, the four source finding algorithms based on 364 the improved GWO, GWO, PSO, and GA can all plan dif-365 ferent paths and successfully locate the source of the plume 366 at other times and different numbers of iterations, as can be 367 seen from the fitness curves, the four algorithms can optimize 368 at the local optimum solution: as the number of iterations 369 grows they continue to jump out of the local optimum and 370 eventually converge to the global optimum. However, the 371 improved GWO has better performance in plume tracking and 372 localization. Four source-finding algorithms were recorded 373 to track the plume and successfully locate the data under 374