【摘要】 针对开敞式码头系泊作业中护舷撞击能量即时预测问题,提出一种基于大数据的分布式K近邻预测法。阐述传统K近邻算法在预测护舷撞击能量时的主要步骤;介绍MapReduce分布式框架的工作原理,给出K近邻算法在MapReduce计算框架下的实现方法,以及交叉验证确定最佳k值的步骤。对算法进行仿真实验,实验结果表明了不同k值对算法预测结果的影响,选取合适k值进行预测时,算法具有较高的准确性。通过将传统K近邻非参数回归方法与大数据Hadoop分布式集群技术相结合,实现海量数据的护舷撞击能量的有效预测,为系泊作业决策提供技术支撑和决策依据。更多还原

【Abstract】 To immediately forecast the impact energy of fender during the ship moored at open sea terminal,a distributed K-nearest neighbor forecasting method based on BigData was advanced.Four processes of K-nearest neighbor forecasting method were presented for impact energy of fender.The operating principle of MapReduce was presented.K-nearest neighbor method and MapReduce were then combined.The best k-value was found by the means of cross-validation.The proposed algorithm was tested with anolog data.The results show that k-values have different impacts on forecasting results.When using the better k-value,more accurate forecasting results are found.By the combination of the K-nearest neighbor forecasting method and the Hadoop distributed cluster,the impact energy of fender with mass data is efficiently forecasted,providing technical supports and evidences for mooring operation decision-making.更多还原