大连海事大学学报
大連海事大學學報
대련해사대학학보
JOURNAL OF DALIAN MARITIME UNIVERSITY
2010年
1期
65-68
,共4页
海底微地形%预测%最小二乘支持向量机(LS-SVM)%数字高程模型(DEM)
海底微地形%預測%最小二乘支持嚮量機(LS-SVM)%數字高程模型(DEM)
해저미지형%예측%최소이승지지향량궤(LS-SVM)%수자고정모형(DEM)
seabed micro-topography%forecasting%least square support vector machine(LS-SVM)%digital elevation model (DEM)
为解决传统机器学习中模型选择、过学习与局部极小值问题,针对自然海底微地形强烈的非线性、不确定性特点,提出海底微地形的LS-SVM预测模型.采用等式约束代替不等式约束,将二次规划问题转化为求解一次线性方程组,提高了收敛速度.实验结果表明,该方法预测结果误差较小,且实时性较好,可满足建立钴结壳最佳切削深度模型的需要.
為解決傳統機器學習中模型選擇、過學習與跼部極小值問題,針對自然海底微地形彊烈的非線性、不確定性特點,提齣海底微地形的LS-SVM預測模型.採用等式約束代替不等式約束,將二次規劃問題轉化為求解一次線性方程組,提高瞭收斂速度.實驗結果錶明,該方法預測結果誤差較小,且實時性較好,可滿足建立鈷結殼最佳切削深度模型的需要.
위해결전통궤기학습중모형선택、과학습여국부겁소치문제,침대자연해저미지형강렬적비선성、불학정성특점,제출해저미지형적LS-SVM예측모형.채용등식약속대체불등식약속,장이차규화문제전화위구해일차선성방정조,제고료수렴속도.실험결과표명,해방법예측결과오차교소,차실시성교호,가만족건립고결각최가절삭심도모형적수요.
To solve such problems as model selecting, over-fitting and local minimum, etc, in traditional machine learning, a least square support vector machine (LS-SVM) based forecasting model of seabed micro-topography was put forward due to the strong non-linear and uncertain characteristics of natural micro-topography in the seabed. Equality constraints were used to re-place inequality constraints, and quadratic programming problem was transformed to solve linear equations , which improved con-vergence rate. Tests show that the proposed method has small prediction error and real time properties, which can meet the requirements of the beat cutting depth model of cobalt crusts.