High-dimensional railway vertical alignment optimization using hybrid differential evolution and gradient descent

Abstract

Designing railway vertical alignment is challenging due to complex geometric constraints, elevation features, and cost savings expectations. Therefore, this paper proposes a Hybrid Vertical Railway Alignment Optimization (HVRAO) model to produce vertical alignment in high dimensions; the proposed model employs a parallel Differential Evolution (DE) algorithm and a swift gradient descent (GD) algorithm in turn, and a subgrade surrogate that utilizes a radial basis function is also proposed to avoid the large subgrade interpolation. Supported by a comprehensive strategy, the HVRAO model can effectively produce stable and optimal vertical alignment. Furthermore, the case study demonstrates that it is capable of creating a railway vertical alignment spanning 52 km with 56 decision variables in a matter of seconds. Finally, the proposed HVRAO framework is also applicable to horizontal alignment optimization and future bilevel alignment models in railway projects.