Metaheuristic Parameter Estimation - Using Parallel Computing and MPI
MetadataVis full innførsel
Arguably, bats have the most sophisticated powered flight mechanism among animals. By mimicking their low-frequency dynamics, significant improvements can be made in energy efficiency, reliability and flexibility of autonomous flight. High-speed videography is now dramatically improving insight into how animals fly by feeding parametric models with large amounts of data. However, accurate estimation of model parameters from experimental data is no trivial task, and most researchers agree that the inverse problem of parameter estimation from time series typically suffers from abundant local optima. Exhaustive methods for solving such problems may require days, weeks or even months to find good solutions, motivating the need for global optimization techniques. In this domain, metaheuristics have shown promising results in recent years. We implement a state-of-the-art metaheuristic, Particle Swarm Optimization (PSO), which is adapted to tackle different problem characteristics. Finally, we show that this method can estimate the parameters of a system of ordinary differential equations with promising signs of convergence, namely the bat model. With ever-increasing access to data and parallel computing hardware, the principle and parallelism of PSO are especially attractive. To this end, the algorithm is parallelized using the Message Passing Interface (MPI), a state-of-the-art implementation of host-to-host communication. Because of its wide applicability and the portability of MPI, the parameter estimation scheme implemented herein is applicable to many fields and models, not only biology but also medicine, industry, finance and more.