Methane plume localization with enhanced self-best reduction and Gaussian improved particle swarm optimization (GiPSO)

Liew, Jia Hun (2024) Methane plume localization with enhanced self-best reduction and Gaussian improved particle swarm optimization (GiPSO). Masters thesis, Sunway University.

Preview

Text Final Hardbound Thesis.pdf - Submitted Version Download (1MB) | Preview

Abstract

Swarm intelligence is a branch of artificial intelligence that studies the collective behavior of groups of social animals such as birds, fish, and bees. It has been used to solve various dynamic problems, including gas leak detection in drone-based leak detection platforms. However, gas plume dispersion has dynamical characteristics often influenced by external environmental factors such as wind direction, wind speed, dispersion rate, and gas density. To investigate the adaptation of swarm intelligence with dynamic modification to further enhance its capability to optimize gas plume dispersion. The research focuses on three questions to enhance the drone swarm optimization algorithm. These three questions steer the research in three separate domains, which helps the evaluation of the performance of our research. The research question, problems, and objectives will be the research directed toward modifying Particle Swarm Optimization (PSO), namely Gaussian improved Particle Swarm Optimization (GiPSO). Firstly, how can swarm intelligence aid in engaging dynamically challenging optimization problems such as gas plume dispersion? To investigate this, our research will investigate the adaptation of the Gaussian gas plume in the simulation. Adapting the Gaussian gas plume model in the simulation provides the experiment with a realistic optimization problem for GiPSO to optimize in the simulation, where we can test the engagement of dynamically challenging optimization problems such as gas plume dispersions. Secondly, our research questions how the Gaussian gas plume model can address the adaptation of swarm intelligence in drone-based gas leakage detection. To address swarm intelligence adaptation in drone-based gas leakage detection, we investigate the existing swarm intelligence capability in optimizing dynamical problems in gas plume detection. Our research employs Gaussian improved Particle Swarm Optimization (GiPSO), derived from modifications implemented on Particle Swarm Optimization (PSO) with Z-axis coefficient clamping and Self-Best reduction mechanism. Z-axis coefficient Clamping provides safety and reduction of drone swarm controlled by GiPSO risk, with the physical collision with the petroleum refinery exhaust. Finally, the third question of our research is how the gas leakage detection algorithm’s performance can be improved when the drone population is low. This guides the research investigating how population growth can impact GiPSO in Optimising Dynamic Problems. To enhance the performance of the population study in GiPSO, the GiPSO self-best reduction mechanism allows GiPSO to re-disperse the swarm when the same particle has been retained as the global best, as it achieves the limitation controlled by the operator. The highlight of our algorithm, GiPSO, exhibits improvement in optimizing the source of leakage in high precision Objective Function Value (OFV). As the experiment setup benchmark specification of DJI Phantom 4 available flight time, GiPSO shows improvement with high success in localizing the source of leakage with population performance peak with 14 particles used in the drone swarm. These further answer our third research question concerning the performance of GiPSO with a low particle population.

Actions (login required)

View Item