Inversion of Time-Evolving Sound Speed Profiles by DREAM-zs With QPSO Proposal Distribution

Abstract

Sound speed profile (SSP) inversion using acoustic data is an efficient approach for estimating SSPs, though it presents a nonlinear and non-Gaussian problem. To improve the spatial resolution of SSP inversion, a range-dependent environment is considered, which increases the number of parameters to be estimated. These challenges limit the performance of commonly used methods, such as the ensemble Kalman filter (EnKF) and particle filter (PF). While EnKFs can handle high-dimensional problems, they assume Gaussian probability distributions. PFs are better suited for highly nonlinear and non-Gaussian systems but are generally more computationally intensive. DiffeRential evolution adaptive Metropolis (DREAM)-zs, a Markov chain Monte Carlo method, is effective for approximating high-dimensional probability distributions, although its convergence speed decreases rapidly as the dimensionality of the parameter space increases. Quantum-behaved particle swarm optimization (QPSO), which combines principles from quantum mechanics and PSO, is a probabilistic optimization algorithm that has been proven successful in various optimization problems. To enhance the efficiency of DREAM-zs, the QPSO proposal distribution is embedded during the burn-in period, forming the DREAM-zqs method. The proposed method is used to track time-evolving SSPs in a range-dependent environment. The inversion problem is formulated in a state-space form, integrating data from the regional ocean modeling system to improve efficiency by shifting the state vector. Simulations and experimental results demonstrate that both DREAM-zs and DREAM-zqs outperform EnKF and PF, with DREAM-zqs achieving faster convergence than DREAM-zs while maintaining inversion accuracy.