Multi-parameter inversion of coupled stress–seepage–damage geotechnical models using a hybridized differential evolution–particle swarm optimization framework

Abstract

Geotechnical back-analysis of multi-physical fields faces mathematical challenges due to the high nonlinearity, non-convexity, and potential ill-posedness of objective functions mapping parameter interactions. This paper addresses the problem of simultaneously estimating geomechanical and hydraulic parameters within a non-linear stress–seepage–damage constitutive model. To mitigate the risks of premature convergence and local optima entrapment common to isolated metaheuristics, we present a hybridized optimization framework combining Differential Evolution (DE) and Particle Swarm Optimization (PSO). The framework utilizes a parallel dual-population topology governed by a stagnation-triggered information-exchange mechanism. The convergence dynamics and stability of the hybridized DE-PSO solver are evaluated over 30 independent realizations against multi-modal test functions, demonstrating competitive performance, particularly in terms of peak precision and standard deviation profiles, alongside state-of-the-art baselines including CMA-ES and SHADE. A multi-parameter back-analysis interface is compiled in C++, establishing a direct link with an implicit finite-element forward engine. The framework is evaluated using a five-dimensional (5D) inversion vector encompassing elastic modulus, Poisson’s ratio, cohesion, friction angle, and hydraulic conductivity. Sensitivity testing under a 5% zero-mean white Gaussian noise contamination indicates that the framework can successfully navigate complex error surfaces under these specific stochastic conditions. This study provides a controlled synthetic numerical evaluation for multi-variable parameter identification, offering a competitive computational approach for coupled environmental geotechnics.