摘要
为解决航材消耗预测影响因素较多、时间相关性复杂、实际消耗样本少等问题,鉴于现有方法难以捕捉小样本航材消耗数据在时间顺序上的逐步依赖关系,提出了基于改进生成对抗网络生成模型和循环神经网络预测模型的航材消耗预测方法。该方法将自回归学习有监督训练与对抗学习无监督训练相结合,进一步挖掘航材消耗数据的潜在静态分布和时态逐步依赖分布,从而生成更加接近真实数据的样本,达到扩充样本量、提高航材消耗预测精度的目的。以某型国产民机部件消耗数据为案例,利用降维可视化方法和预测分数评价模型的生成性能和预测精度。通过与其他方法对比,所提方法预测值的平均绝对误差减少了3%,验证了所提方法解决航材预测问题的有效性。
In order to solve the problems of many factors affecting spare parts consumption forecast,complex time correlation,and few actual consumption samples,in view of the difficulty of capturing the gradual dependence of small sample spare parts consumption data in time sequence with existing methods,a spare parts consumption prediction method based on the improved generation adversarial network generative model and the recurrent neural network prediction model is proposed.This method combines the supervised training of autoregressive learning with the unsupervised training of adversarial learning,and further explores the potential static distribution and temporal step-by-step dependence distribution of the spare parts consumption data.Samples that are closer to the real data are generated to achieve the purpose of expanding the sample volume and improving the forecast accuracy of spare parts consumption.Taking the consumption data of a certain type of domestic civil aircraft as an example,the generation performance and forecast accuracy of the model are evaluated by using the dimensionality reduction visualization method and the forecast score.Compared with other methods,the average absolute error of the predicted value presented by the proposed method is reduced by 3%,which verifies the effectiveness of the proposed method in solving the problem of spare parts prediction.
作者
曾浩然
冯蕴雯
路成
潘维煌
ZENG Haoran;FENG Yunwen;LU Cheng;PAN Weihuang(School of Aeronautics,Northwestern Polytechnical University,Xi’an 710072,China)
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2023年第10期3132-3138,共7页
Systems Engineering and Electronics
基金
国家自然科学基金(51875465)资助课题。
关键词
国产民机
航材消耗预测
时间序列
生成对抗网络
domestic civil aircraft
spare parts consumption forecast
time sequence
generative adversarial network(GAN)