Machine Learning-based optimization of literature models for submarine cable installation in sandy seabeds

Offshore wind farms are rapidly emerging as a key solution towards green energy production. Offshore infrastructures are connected to onshore substations via submarine cables, which are usually buried in trenches for protection. Submarine cables also serve a strategic role as data traffic infrastructure, highlighting the need for efficient and optimized installation designs. Cable installation techniques vary with geotechnical conditions, with ploughing widely preferred due to its versatility across different soil types. Reliable predictions of the tow force needed to drive the plough is critical for optimal cable installation planning. The literature presents several analytical approaches to estimate this force, based on earth pressure theories. In the case of granular seabed, the tow force comprises a static component and a velocity-dependent (dynamic) component. The static component arises from (i) friction between the plough and the seabed, and (ii) soil resistance (Lauder, 2010). In contrast, the nature of the dynamic component remains debated, but is commonly attributed to shear-induced pore water pressures under partially drained conditions (see e.g. Palmer, 1999). These formulas require a set of semiempirical factors which have been calibrated from small-scale laboratory tests (Lauder, 2010; Robinson et al., 2017). However, when applied to real-world field projects, their predictive capabilities have shown significant limitations (Marveggio et al, 2025). To overcome these limitations, Machine Learning (ML)-based modeling has recently emerged as a promising alternative for analyzing complex nonlinear processes in various geotechnical engineering fields. This paper explores the possibility of using regression-based algorithms trained on real-world project data to recalibrate existing formulations, popular among practitioners.