基于机器学习方法的航空消耗件需求自适应预测  

Adaptive Demand Prediction of Aviation Consumable Spare Parts Based on Machine Learning Method

作　　者：付维方[1] 穆彩虹 刘英杰 FU Wei-fang;MU Cai-hong;LIU Ying-jie(School of Aviation Engineering,Civil Aviation University of China,Tianjin 300300,China;Aircraft Maintenance&Engineering Corporation,Beijing 100621,China)

机构地区：[1]中国民航大学航空工程学院,天津300300 [2]北京飞机维修工程有限公司,北京100621

出　　处：《科学技术与工程》2022年第11期4609-4617,共9页Science Technology and Engineering

基　　金：中国民航大学中央高校基金(3122016D010)。

摘　　要：企业状态不稳定性、消耗件故障规律不确定性及需求特征的动态性,特别是新冠疫情期间航空公司航班的大量停飞和逐渐恢复导致固定单一需求预测方法存在较大偏差。基于平均绝对误差和均方误差进行不同需求模式预测方法筛选,构建自组织特征映射网(self-organizing feature map,SOFM)对需求时间序列聚类,提出不同聚类模式和预测方法映射关系并实现数据与方法动态自适应。此自适应预测框架能够实现不同航材需求模式识别、多预测方法决策及同一航材多阶段动态预测。通过某航空公司实例验证表明该自适应框架具有较好的应用效果,适用于各类型的消耗备件需求预测。The instability of enterprise status,the uncertainty of failure law of consumable spare parts and the dynamic demand characteristics,especially a large number of flights are grounded and gradually resume during the outbreak,lead to the large deviation in the fixed single demand prediction methods.Based on the mean absolute deviation and mean square deviation,prediction methods of different demand pattern were screened to construct SOFM neural network to cluster demand time series.The mapping relationship between different clustering modes and prediction methods was proposed,and the dynamic adaptation of data and methods was realized.This adaptive prediction framework can realize the recognition of different demand pattern,multi-prediction method decision-making and multi-stage dynamic prediction of the same spare part.The example of airlines shows that the adaptive framework has good application effect to forecast the demand of different classification of consumable spare parts.

关 键 词：SOFM神经网络 需求模式识别 动态预测 航空消耗件 

分 类 号：V267[航空宇航科学与技术—航空宇航制造工程]