A prediction method for deck-motion based on online least square support vector machine and genetic algorithm

作者：Liu, Xixiang; Wang, Qiming; Huang, Rong; Wang, Songbing; Liu, Xianjun

来源：Journal of Marine Science and Technology (Japan), 2019, 24(2): 382-397.

DOI：10.1007/s00773-018-0557-z

• Prediction for deck-motion
• Online LSSVM
• Information accumulation
• window stage
• GA
• Fitness function

The prediction accuracy and prediction time are key elements that determine the landing safety of the carrier-based aircraft when ship motions in six-degree freedom are difficult to restrain. The classical prediction methods, such as auto-regressive and moving average model (ARMA) and radical basis function neural network (RBF-NN), always suffer from the drawbacks of short prediction time and low accuracy caused by the nonlinearity and randomness of deck-motion. Aiming to lengthen prediction time and improve prediction accuracy, an online prediction method based on Least Square Support Vector Machine (LSSVM) is proposed with the comprehensive consideration of the characters of deck-motion and inertial navigation system (INS)an instrument to measure motions including deck-motion. For the sequentiality and timeliness of deck-motion, the proposed online LSSVM prediction method is divided into two stagesinformation accumulation stage and information window stage. To acquire optimal parameters for LSSVMthe number of samples, the length of sample, the parameter for kernel function and the penalty factor, genetic algorithm (GA) is adopted; in addition, the fitness function for GA is designed according to the periodicity of deck-motion. The prediction tests are conducted with data from deck-motion models and sea trail, respectively, and the results indicate that the proposed method can provide a preferable result compared with those methods based on ARMA, RBF-NN and Particle Swarm Optimization and Kernel Extreme Learning Machine (PSO-KELM) when the data mapping relations are nonlinear and changeable.

• 出版日期2019-6
• 单位东南大学