Research on container ship trajectory prediction based on the EMD-PSO-GRU-RBFNN algorithm

Abstract

Ship trajectory prediction is challenging due to its strongly time-varying, non-stationary, and highly nonlinear characteristics. To address this issue, a hybrid prediction framework integrating Empirical Mode Decomposition (EMD), Particle Swarm Optimization (PSO), Gated Recurrent Unit (GRU), and Radial Basis Function Neural Network (RBFNN) is proposed. First, a PSO-RBFNN model generates preliminary predictions of vessel latitude and longitude, and residual sequences are obtained by subtracting the predicted values from the observations. EMD is then applied to decompose the residual sequence into multiple Intrinsic Mode Function (IMF) components with different frequency characteristics, along with a residual component, thereby reducing data complexity and nonlinearity. PSO is employed to optimize key hyperparameters of the RBFNN and GRU models using training and validation data, and the optimized configuration is fixed and reused for modeling all decomposed subsequences. Each component is independently modeled and predicted, and the final vessel trajectory is reconstructed by aggregating the predicted components. Validation using Automatic Identification System (AIS) data demonstrates that the proposed EMD-PSO-GRU-RBFNN model achieves strong stability and superior predictive accuracy compared with existing methods. This study provides an effective framework for high-precision ship trajectory prediction and shows considerable potential for practical maritime applications.