Particle swarm optimization has developed as a popular method of solution discovery for many
numerical problems. Yet, these methods are computationally expensive and require numerous runs
to confidently discover a quasi-optimal solution. These major bottlenecks - execution time and
a low probability of optimal solution convergence - restrict the usage of particle swarm methods
for time-sensitive calculations. In an exploration of the finite thrust arc problem as a benchmark
case, this thesis analyzes particle swarm optimization improvements in an effort to broaden the
algorithm’s utility.
The research presented here studies two methods to significantly improve each of the major particle
swarm bottlenecks. First, reducing execution time is explored through various means of problem
implementation. Memory-optimized single-threaded and parallelized C++ algorithms are found to
perform up to 96% faster then MATLAB. Second, minimizing premature convergence is studied
with rehydration - a proposed method of resetting a portion of the swarm under population stagna-
tion. Rehydration results demonstrate up to a 44% improvement in average solution discovery. In
employing these two methods in concert, this research suggests the potential feasibility of reliable,
real-time particle swarm optimization techniques.