Modeling and Optimized Support Vector Regression Algorithm-based Identification of Ship Dynamics
26. Februar 2018, 16:15 , 17:45
This thesis deals with the problem of how to derive a simplistic model feasible for describing dynamic of types of ships. The model should be expressed in a simple form with satisfying accuracy, which can describe ship maneuvering characteristics for the control design.
The problem of deriving ship dynamic model is addressed first with the modification and simplification of a complex vectoral representation in 6 degrees of freedom (DOF). The 6 DOF complicated dynamic model is simplified through several pieces of reasonable assumptions which are exampled as ships moving in horizontal plane in the ideal fluid, the uniformly distribution of ship masses, the port-starboard symmetry, etc. In the process of simplification, the tradeoff between accuracy and possibility of estimation of the simplified model is regarded as the key criteria. Consequently, a 3 DOF dynamic model in simple form used to simultaneously
capture surge motions and steering motions is found.
Solving the problem of deriving ship dynamic model requires the estimation of parameters in the model through an achievable technique. Considering the restricted condition of availably used information for estimation, system identification in combination with free-running model test or full-scale ship trail is undertaken to estimate parameters in the simplified ship dynamic model. Attracted by the merits and demerits of support vector regression algorithm (SVR) shown while applied to parameter estimation in maritime field, SVR is specific to the selected parameter estimation method.
SVR has poor ability of on-line identification and inefficiency in dealing with massive data set, but it demonstrates good generalization ability due to it has the criteria of structural risk minimization, guarantees a globally optimal solution by adopting convex quadratic programming, and can overcome the dimensionality curse thanks to the use of kernel function. Furthermore, due to attentions to applying a small amount of data to off-line identification of the simplified ship dynamic model are paid, the disadvantages of SVR are not barriers to its application in this thesis. To remedy the deficiency of SVR, i.e. the identification results are sensitive to the structural parameters in SVR such as the insensitivity factor, the regularization parameter, and kernel parameters, the artificial bee colony algorithm (ABC) with superior optimization performance rather than the generally used cross-validation method and particle swarm optimization algorithm (PSO) is confirmed to tune parameters in SVR.
Finally, data extracted from a set of steady-state and zigzag maneuvers are applied to test the effectiveness of the simplified ship dynamic model and to validate the performance of the optimized SVR on identifying model. Results show an effective model in a simple form for describing ship dynamics, and also validate the performance of the proposed parameter estimation method.
Betreuer: Prof. Dr.-Ing. Axel Hahn , Prof. Dr. Yuanqiao Wen