acceptodds
Desk rejected at ICLR 2027

Multi-View Parallel Cooperative Optimization for Dynamic Launch Vehicle Attitude Control

Desk rejected at ICLR 2027, so left out of the paper lists and the map.

Abstract

Dynamic Launch Vehicle Attitude Control (DLVAC) involves designing correction networks across dynamic flight stages. In practice, this poses two key challenges: (i) exploring strongly constrained design spaces and (ii) transferring reusable design knowledge. Several Dynamic Optimization Algorithms (DOAs) can address strongly constrained problems to some extent. However, due to the need for numerous iterations and their tendency to become trapped in local optima, they are somewhat inadequate when addressing strongly constrained optimization problems. To address the limitations of existing works, we propose a novel approach called Multi-View Parallel Cooperative Optimization (MPCO). MPCO constructs multiple subprocesses to carry out the optimization in parallel, enhancing optimization efficiency. In the iterative process, subprocesses gather problem-relevant information from multiple perspectives, adjusting search directions to enhance the exploratory capability of MPCO. Additionally, cooperation among subprocesses is facilitated through a knowledge transfer strategy, which helps reduce the issue of becoming trapped in local optima. Using a set of common benchmark functions and a set of practical launch vehicle control parameter design tasks, we compare the performance of MPCO with nine advanced algorithms. MPCO achieves superior and feasible solutions in less time, demonstrating excellent efficiency and stability.

Primary area:
applications to physical sciences (physics, chemistry, biology, etc.)
Keywords:
Constraint Optimization, Dynamic Launch Vehicle Attitude Control, Dynamic Optimization Algorithms

Related papers

Discussion (0)

Sign in to comment.