摘要
针对海浪水压场的短时平稳性,采用局域支持向量机进行预测滤波。首先分析了海浪水压场信号的混沌特性,并根据其混沌特性,提取训练空间中与当前相点的行为特征密切相关的最近邻点作为训练样本对支持向量机进行训练,减少了训练样本的数目,节省了网络学习时间,从而可实时对网络参数进行更新,使支持向量机能够跟随海浪的变化。实际计算表明这种算法能够以较快的学习速度和较高准确度实现海浪预测,能够克服由于海浪的短时平稳性所带来的随时间的增长预测精度下降的问题。
Duo to the short-time stability of wave hydrodynamic pressure, a prediction filtering method of local support vector machine （LSVM） was proposed. Firstly the chaos character of wave hydrodynamic pressure signal was analyzed, and then the nearest samples which were intimately connected with the current phase point in behavioural trait were extracted from the training space to train the SVM. Then the number of training samples was reduced, and it saved the net learning time. So the SVM net parameters could be real-timely renewed and made the SVM fellow the wave changing. The calculation results show the method can realize the accurate fast prediction with short training time. At the time, it can overcome the prediction accuracy descent for the un-stability ascent with the time escaped.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2008年第23期6470-6472,6476,共4页
Journal of System Simulation
基金
国防重点实验室基金项目资助(51444060101JB1108)
关键词
海浪水压场
支持向量机
混沌预测
短时平稳
wave hydrodynamic pressure
support vector machine
chaos prediction
short-time stability
