摘要
针对舰船装备维修费样本量小、不确定性大,且受最近一次维修费影响大的特点,将马尔科夫链引入到舰船装备维修费组合预测建模中。针对马尔科夫链状态空间划分的主观性,采用k均值聚类算法对各单项预测模型拟合误差进行聚类,形成k个状态空间;针对以误差最小为准则确定组合预测权重可能导致过拟合的问题,从拟合误差的准确性和发展趋势的相似性两个角度,分别给出了舰船装备维修费组合预测权重确定方法,并结合马尔科夫链对组合预测模型进行了修正。最后,实例验证了该方法的可行性和有效性。
The problems of small sample size and great uncertainty often exist in ship equipment maintenance cost.The maintenance cost is impacted largely by the lasting maintenance cost.Markov chain is introduced to combination forecast model construction for ship equipment maintenance cost.Due to the subjectivity in dividing state space of Markov chain,the k\|means clustering analysis of individual forecasts is done based on similarity in their fitting error,and then the k state spaces are forming.Given that the minimum error criterion in determining the weights of combination forecast will result in overfitting,methods for determining weights of combination forecast are given respectively from both the views of development trend similarity and the fitting precision.Next,the combination forecast model is revised by Markov chain.Finally an example is given to verify the feasibility and effectiveness of this method.
作者
蒋国萍
谢力
孙胜祥
安天霞
尚凡葵
JIANG Guo-ping;XIE Li;SUN Sheng-xiang;AN Tian-xia;SHANG Fan-kui(Dept.of Management Engineering and Equipment Economics,Naval Univ.of Engineering, Wuhan 430033,China;Repair Plant for Navigation Safeguard Equipment in Qingdao,Qingdao 266000,China;Navy Military Representative Office,Qingdao 266000,China)
出处
《海军工程大学学报》
CAS
北大核心
2018年第2期13-17,共5页
Journal of Naval University of Engineering
关键词
马尔科夫链
组合预测
小样本
舰船装备维修费
Markov chain
combination forecast
small sample
ship equipment maintenance cost
