基于Mini-batch神经网络的船舶柴油机风险等级预测  

Prediction of marine diesel engine risk level based on mini-batch neural network

作　　者：尚前明[1] 王潇 曹召 刘治江 邓晓光[1] 

机构地区：[1]武汉理工大学能源与动力工程学院,湖北武汉430063

出　　处：《中国修船》2018年第4期35-40,共6页China Shiprepair

基　　金：国家自然科学基金(51579200)

摘　　要：文章针对现有的船舶柴油机健康管理智能化程度不高、管理技术落后等问题,引入神经网络风险预测方法。提出Mini-batch梯度下降方法,通过在网络训练不同阶段使用不同权重的训练样本,提升原有全量梯度下降算法预测模型不准确和较难更新等问题。并从船舶柴油机在一次机务风险所承担的风险着手,建立设备健康风险状态的综合评价指标体系,通过Minibatch梯度下降优化了风险等级分类器。结果表明,该方法能够对柴油机风险等级进行有效预测。Aiming at the problems of low intelligence and backward management technology of marine diesel engine health management,a neural network risk prediction method is introduced in this paper. The Mini-batch gradient descent method is proposed. By using training samples with different weights in different stages of network training,the prediction model of the original total gradient descent algorithm is inaccurate and difficult to update.Based on the risk of marine diesel engine in primary engine maintenance,a comprehensive evaluation index system of equipment health risk state is established,and the risk grade classifier is optimized by Mini-batch gradient descent. The results show that the proposed method can be applied to diesel oil. The risk level of the machine is predicted effectively.

关 键 词：船舶柴油机 Mini-batch梯度下降 风险等级预测 神经网络 

分 类 号：U672[交通运输工程—船舶及航道工程]