An Optimized Thrust Allocation Algorithm for Dynamic Positioning System Based on RBF Neural Network

Abstract

Abstract The thrust allocation of Dynamic Positioning System (DPS) equipped with multiple thrusters is usually formulated as an optimization problem. Hydrodynamic interaction effects such as thruster-thruster interaction results in thrust loss. This interaction is generally avoided by defining forbidden zones for some azimuth angles. However, it leads to a higher power consumption and stuck thrust angles. For the purpose of improving the traditional Forbidden Zone (FZ) method, this paper proposes an optimized thrust allocation algorithm based on Radial Basis Function (RBF) neural network and Sequential Quadratic Programming (SQP) algorithm, named RBF-SQP. The thrust coefficient is introduced to express the thrust loss which is then incorporated into the mathematical model to remove forbidden zones. Specifically, the RBF neural network is constructed to approximate the thrust efficiency function, and the SQP algorithm is selected to solve the nonlinear optimization problem. The training dataset of RBF neural network is obtained from the model test of thrust-thrust interaction. Numerical simulations for the dynamic positioning of a semi-submersible platform are conducted under typical operating conditions. The simulation results demonstrate that the demanded forces can be correctly distributed among available thrusters. Compared with the traditional methods, the proposed thrust allocation algorithm can achieve a lower power consumption. Moreover, the advantages of considering hydrodynamic interaction effects and utilizing a neural network for function fitting are also highlighted, indicating a practical application prospect of the optimized algorithm.