| Odor source localization(OSL)is the kernel problem that should be tackled in the first place in odor/gas-related rescue and inspection applications,however,traditional OSL methods have the drawbacks such as high safety risk,low accuracy etc.Therefore,this dissertation investigates OSL methods based on multiple position fixed wireless sensor nodes and multiple mobile sensor nodes,respectively.The former tackle OSL problems in the scenarios where airflow is relatively stable and long-term monitoring is needed,while the latter the scenarios containing fiercely fluctuating airflow.The main research achievements can be concluded as follows:Ranking-fitness-based genetic algorithm(RFGA)is used with least-squares estimation for OSL using a wireless sensor network(WSN).Simulation results show that,for solving the least-squares formulation of the WSN based OSL,simple genetic algorithm is not convergent,and the traditional single-point-based search methods can hardly overcome the local convergent problem,while the RFGA is globally convergent and capable of accurately estimating the location and release rate of the odor source,as well as the environmental diffusion coefficient,at the same time.Two random-walk-based methods are proposed for plume finding with multiple mobile sensing nodes,and an artificial potential field mechanism is then incorporated into the proposed methods to calculate the turning angle of individual walking steps.Compared with the counterpart methods,the proposed methods have the merit of not having to rely on the real-time wind speed.Simulation results show that the proposed method can efficiently realize plume finding,and that the incorporated artificial potential field mechanism is beneficial to steer different nodes to different areas,thereby improving the efficiency of plume finding.A method based on online reinforcement learning is proposed for plume tracking with multiple mobile sensing nodes,and moreover,collaborative reinforcement learning is realized by sharing the experience of the sensing nodes.Experimental results show that,compared with the existing offline-optimization-based counterpart methods,the proposed method has higher adaptability to dynamic airflow environments.In addition,the experience sharing mechanism improves the success rate of plume tracking.A behavior-based method framework is proposed for OSL using multiple mobile sensing nodes,and a general scheme is also proposed for implementing evolutionary computation(EC)in the plume exploitation behavior of the proposed method framework.Based on the proposed framework and scheme,using artificial bee colony algorithm and genetic algorithm as two specific EC methods,two different EC-based methods are proposed for OSL using multiple mobile sensing nodes.Simulation results show that the proposed framework can select different behaviors according to different environmental conditions for controlling the mobile sensing nodes,the proposed scheme can successfully realize collaboration among multiple sensing nodes with the collaboration mechanism embodied in the adopted EC method,and the two proposed EC-based OSL methods can locate the odor source with considerably high accuracy and time-efficiency. |
