Timothy Guy Thompson

Parallel Evolutionary Multi-Objective Optimization on Large, Heterogeneous Clusters: An Applications Perspective

Real-world operational use of parallel multi-objective evolutionary algorithms requires successful searches in constrained wall-clock periods, limited trial-and-error algorithmic analysis, and scalable use of heterogeneous computing hardware. This study provides a cross-disciplinary collaborative effort to assess and adapt parallel multi-objective evolutionary algorithms for operational use in satellite constellation design using large dedicated clusters with heterogeneous processor speeds/architectures. A statistical, metric-based evaluation framework is used to demonstrate how time-continuation, asynchronous evolution, dynamic population sizing, and epsilon dominance archiving can be used to enhance both simple master–slave parallelization strategies and more complex multiple-population schemes. Results for a benchmark constellation design coverage problem show that simple master– slave schemes that exploit time-continuation are often sufficient and potentially superior to complex multiple-population schemes.

Pareto-hypervolumes for the Reconfiguration of Satellite Constellations

A satellite constellation is designed to perform its mission with a nominal number of spacecraft. When a reduction in capacity is experienced, for whatever reason, the remaining constellation may be able to restore performance through reconfiguration. In this work we present a general framework that exploits recent efforts in parallel multiobjective evolutionary computation, to reconfigure satellite constellations that have suffered the loss of one or more of their vehicles. The framework is illustrated through several loss scenarios for the Global Positioning System constellation. Pareto-hypervolumes are constructed which are the set of solutions that approximate the optimum tradeoff between minimizing cost and risk while maximizing performance. The decision making processes using the high-dimensional data sets is illustrated. The results demonstrate a pragmatic approach to optimization wherein the insights gained from a multi-objective view of the design space tradeoffs allow for informed decision making. Nomenclature

A Framework for the Discovery of Passive-Control, Minimum Energy Satellite Constellations

The hypothesis that high fidelity astrodynamics simulations that account for orbital perturbations can be exploited to discover passive control and/or minimum energy satellite constellations to dramatically reduce propellant demands and improve coverage performance, is explored. A new framework that couples many-objective evolutionary algorithms, high fidelity astrodynamics simulation, massive parallel computing, and visual analytics, is introduced and demonstrated to revisit Draim’s original 4-satellite global coverage problem. Draim’s original constellation design has a 60% coverage deterioration due to orbital perturbations over a 10 year period. A significant finding of this study is that this deterioration can be effectively eliminated through the discovery of constellation configurations that are capable of using perturbing accelerations instead of propellant to restore global coverage performance over decadal time scales. Further, a generalized stationkeeping model is contributed that when incorporated into the framework allows for a complete characterization of the performance tradeoffs that emerge when using passive control to seek minimum energy designs. With the investment of over 21 million hours of compute time, the results are drawn from one of the largest computational experiments to-date for the design of satellite constellations. Overall this study contributes a new paradigm for the way we conceive, design, and populate satellite constellations.

Constellation Design with Parallel Multi-Objective Evolutionary Computation

‡Multi-objective evolutionary algorithms (MOEA) have been shown to be effective optimization tools to search the complex trade-off space s of satellite constellation design. Often the metrics that make up the design trade-off req uire lengthy function evaluation time, resulting in a decreased utility of serial MOEA. In this research the authors implement two parallel processing MOEA paradigms, the master-slave and island models, on a heterogeneous system of processors and operating syst ems. The efficiency and effectiveness of each approach is studied in the conte xt of a regional coverage design problem. The island scheme outperforms the master-slave model with respect to efficiency. A study of the search dynamics for each paradigm demonstrates that both reliably meet the goals of multi-objective optimization (progressing towar ds the Pareto-optimal front while maintaining a diverse set of solutions). A key conclu sion of this research is that both paradigms provide excellent approximations of the true Pareto frontier using a single seed, and when combined across multiple trial runs find nearly the entire set of Pareto-optimal solutions.