%0 Journal Article
%T 基于ASAE深度学习预测海洋气象对船舶航速的影响<br>Prediction of marine meteorological effect on ship speed based on ASAE deep learning
%A 王胜正
%A 申心泉
%A 赵建森
%A 冀宝仙
%A 杨平安
%J 交通运输工程学报
%D 2018
%X 为了有效地预测海洋气象对船舶航速的影响,在稀疏自编码(SAE)网络模型的基础上提出交替稀疏自编码(ASAE)网络模型; 构建了海洋气象对船舶航速影响的预测框架,利用关联规则方法对航行数据进行特征选择,挖掘了船速影响因素及其隐含关系; 整合了中国远洋海运集团有限公司提供的船舶航行数据以及美国国家海洋和大气管理局提供的气象数据,用训练样本对ASAE网络模型进行训练,用测试样本对ASAE网络模型进行验证,并与支持向量回归(SVR)模型、反向传播神经网络(BPNN)模型、深度信念网络(DBN)模型及SAE网络模型的预测结果进行了对比。研究结果表明:ASAE网络模型的训练时间和海洋气象对船舶航速影响预测值的均方根误差分别为8.2 s和0.287 3 kn,与SVR模型、BPNN模型、DBN模型及SAE网络模型相比,训练时间分别缩短了1 683.1、66.9、2.0、1.5 s,预测准确度分别提高了0.045 5、0.296 9、0.153 4、0.178 6 kn; ASAE网络模型的预测结果更符合实际海况,可动态掌握海洋气象对船舶航速的影响; 通过预测的航速影响值来推算实际航速可为气象导航优化船舶运输过程起到辅助作用,在进行航线规划、航速推荐等航行优化策略时能准确考虑海洋气象所产生的复杂影响,从而改善船舶运营能效指标,实现节能、低碳、绿色航行的宗旨。<br>In order to effectively predict the marine meteorological effect on ship speed, the alternating sparse auto-encoders(ASAE)network model based on the sparse auto-encoders(SAE)network model was proposed. A framework for predicting the marine meteorological effect on ship speed was constructed, and the association rules method was proposed to feature the navigation data, so as to excavate the factors and implicit relations affecting ship speed. Through the integration of ship navigation data provided by China COSCO Shipping Group and the meteorological data provided by the National Oceanic and Atmospheric Administration, ASAE network model was trained with training samples and verified with test samples, and the prediction result was compared to the results gained by support vector regression(SVR)model, back propagation neural network(BPNN)model, deep belief network(DBN)model and SAE network model. Research result shows that the training time and the mean squared error of marine meteorological effect on ship speed gained by ASAE network model are 8.2 s and 0.287 3 kn, respectively. Compared to SVR model, BPNN model, DBN model, and SAE network model, ASAE network model can shorten the training time by 1 683.1, 66.9, 2.0 and 1.5 s, respectively, and can increases the prediction accuracy by 0.045 5, 0.296 9, 0.153 4 and 0.178 6 kn. The forecast result of ASAE network model is more in line with actual sea condition, and can dynamically master the marine meteorological effect on ship speed. Estimating the actual speed through the predicted values can optimize the ship transportation process in meteorological navigation. It plays an auxiliary role to accurately consider the complex impacts of marine meteorological on navigation optimization strategies such as route planning and speed recommendation. Thereby, it improves the energy efficiency indicator of ship operation and achieves the purpose of energy saving, low-carbon and green navigation. 5 tabs, 6 figs, 23 refs
%K 交通信息工程
%K 智能航行
%K 船舶航速
%K 深度学习
%K 交替稀疏自编码
%K 关联规则
%K 气象因子&lt
%K br&gt
%K traffic information engineering
%K intelligence voyage
%K ship speed
%K deep learning
%K alternating sparse auto-encoder
%K association rule
%K meteorological factor
%U http://transport.chd.edu.cn/oa/DArticle.aspx?type=view&id=201802015