基于改进LSTM模型的航空安全预测方法研究

doi:10.12305/j.issn.1001-506X.2022.02.26

摘要/Abstract

摘要：

精确的航空安全预测是科学开展安全预警的前提。航空事故不仅致因机理复杂, 还存在迟滞效应, 给安全样本时序信息的深度挖掘加大了难度。基于此, 提出一种基于改进长短期记忆(long short-term memory, LSTM)模型的航空安全预测新方法。首先基于相关系数热图优选致因指标, 再以步进搜索和Adam算法相结合的方式优化LSTM模型超参数, 最后以2019年某型运输机事故数据为算例, 选取多种常用时序预测模型作为对照。实验结果表明本文所提方法, 预测误差较现有方法降低了28%以上, 同时具有较好的泛化能力和鲁棒性。

关键词: 航空安全, 神经网络, 长短期记忆, 堆叠式, 多步预测

Abstract:

Accurate aviation safety prediction is the premise of efficient safety early warning. Aviation accidents are not only caused by complex mechanism, but also have hysteresis effect, which makes it more difficult to excavate the time-series information of safety samples in depth. Based on this, an aviation safety forecast method based on improved long short-term memory (LSTM) model is proposed. Firstly, cause indicators are optimized based on the correlation coefficient heat maps. Then the model parameters of LSTM model are optimized by the step search and Adam algorithm. Finally, taking the accident data of a certain transport aircraft in 2019 as an example, select a variety of time-series forecasting model commonly used as a control. According to the experimental results, compared with the existing methods, the prediction error of the proposed method can be reduced by more than 28%, with excellent generalization ability and robustness.

Key words: aviation safety, neural network, long short-term memory (LSTM), multi-layers, multi-step prediction

中图分类号:

V328

曾航, 张红梅, 任博, 崔利杰, 武江南. 基于改进LSTM模型的航空安全预测方法研究[J]. 系统工程与电子技术, 2022, 44(2): 569-576.

Hang ZENG, Hongmei ZHANG, Bo REN, Lijie CUI, Jiangnan WU. Aviation safety prediction method research based on improved LSTM model[J]. Systems Engineering and Electronics, 2022, 44(2): 569-576.

图/表 13

图1

图2

图3

图4

表1

图5

表2

图6

表3

表4

图7

表5

图8

参考文献 31

1	国家统计局. 2010—2019年旅客周转量数据统计[EB/OL]. [2021-04-01]. https://data.stats.gov.cn/easyquery.htm?cn=C01&zb=A0G05&sj=2019.
2	DI G , PATRIARCA R , MANCINI M , et al. Overall safety performance of air traffic management system: forecasting and monitoring[J]. Safety Science, 2015, 72 (2): 351- 362.
3	LUKÁČOVÁ A, BABIČ F, PARALIČ J, et al. Building the prediction model from the aviation incident data[C]//Proc. of the IEEE 12th International Symposium on Applied Machine Intelligence and Informatics, 2014: 365-369.
4	LI C S , CHIANG T W . Complex neurofuzzy ARIMA forecasting—a new approach using complex fuzzy sets[J]. IEEE Trans.on Fuzzy Systems, 2013, 21 (3): 567- 584. doi: 10.1109/TFUZZ.2012.2226890
5	甘旭升, 端木京顺, 高建国, 等. 基于ARIMA模型的航空装备事故时序预测[J]. 中国安全科学学学报, 2012, 22 (3): 98- 102.
	GAN X S , DUANMU J S , GAO J G , et al. Time series prediction of aviation equipment accident based on ARIMA model[J]. China Safety Science Journal, 2012, 22 (3): 98- 102.
6	杜毅. 基于灰色理论的飞行事故率和事故症候率预测[J]. 中国民航大学学报, 2007, 25 (6): 9- 10. doi: 10.3969/j.issn.1001-5590.2007.06.003
	DU Y . Forecast analysis of flight accident-rate or flight incident-rate[J]. Journal of Civil Aviation University of China, 2007, 25 (6): 9- 10. doi: 10.3969/j.issn.1001-5590.2007.06.003
7	陈农田, 张宏钰, 曹玉宽. 基于灰色马尔可夫理论的进近着陆不安全事件预测[J]. 数学的实践与认识, 2020, 50 (4): 306- 314.
	CHEN N T , ZHANG H Y , CAO Y K . Prediction of approaching landing unsafe events based on grey Markov method[J]. Mathmatics in Practice and Theory, 2020, 50 (4): 306- 314.
8	高建国, 端木京顺, 李闯, 等. 基于模糊均生函数和最优子集回归的飞行事故率预测[J]. 中国安全科学学报, 2011, 21 (6): 43- 47. doi: 10.3969/j.issn.1003-3033.2011.06.007
	GAO J G , DUANMU J S , LI C , et al. Prediction of flight accident rate based on fuzzy mean generating function and optimal subset regression[J]. China Safety Science Journal, 2011, 21 (6): 43- 47. doi: 10.3969/j.issn.1003-3033.2011.06.007
9	GAO Z Y , LI J W , WANG R X , et al. Prognostics uncertainty reduction by right-time prediction of remaining useful life based on hidden Markov model and proportional hazard model[J]. Eksploatacja I Niezawodnosc-maintenance and Reliability, 2021, 23 (1): 154- 164. doi: 10.17531/ein.2021.1.16
10	马春茂, 邵延君, 潘宏侠, 等. 基于灰色马尔可夫模型的装备故障间隔期预测研究[J]. 兵工学报, 2013, 34 (9): 1193- 1196.
	MA C M , SHAO Y J , PAN H X , et al. TBF prediction of equipment based on the gray markov model[J]. Acta Armamentarii, 2013, 34 (9): 1193- 1196.
11	ZHANG X G , MAHADEVAN S . Ensemble machine learning models for aviation incident risk prediction[J]. Decision Support Systems, 2019, 116 (1): 48- 63.
12	ROSA M A V . Prediction of aircraft safety incidents using Bayesian inference and hierarchical structures[J]. Safety Science, 2018, 104 (4): 216- 230.
13	李大伟, 徐浩军, 胡良谋, 等. 基于支持向量机的飞行事故率预测模型[J]. 数学的实践与认识, 2009, 39 (8): 124- 128.
	LI D W , XU H J , HU L M , et al. The prediction modelling of flight accident rate based on support vector machine[J]. Mathematics in Practice And Theory, 2009, 39 (8): 124- 128.
14	端木京顺, 甘旭升, 史超. 基于最小二乘支持向量机的飞行事故预测[J]. 微计算机信息, 2009, 25 (13): 285- 287. doi: 10.3969/j.issn.1008-0570.2009.13.118
	DUANMU J S , GAN X S , SHI C . Flight accident prediction based on least squares support vector machines[J]. Microcomputer Information, 2009, 25 (13): 285- 287. doi: 10.3969/j.issn.1008-0570.2009.13.118
15	于洪霞, 李兴. 一种基于MLP神经网络的大额损失飞行事故预测模型[J]. 上海电力学院学报, 2016, 32 (5): 504- 506. doi: 10.3969/j.issn.1006-4729.2016.05.019
	YU H X , LI X . Analysis and forecast mode of major or loss flight accidents based on mlp neural networks method[J]. Journal of Shanghai University of Electric Power, 2016, 32 (5): 504- 506. doi: 10.3969/j.issn.1006-4729.2016.05.019
16	SUNDERMEYER M , NEY H , SCHLUETER R . From feedforward to recurrent LSTM neural networks for language modeling[J]. IEEE/ACM Trans.on Audio Speech and Language Processing, 2015, 23 (3): 517- 529. doi: 10.1109/TASLP.2015.2400218
17	SALAH B , ALI F , ALI O , et al. Optimal deep learning LSTM model for electric load forecasting using feature selection and genetic algorithm: comparison with machine learning approaches[J]. Energies, 2018, 11 (7): 1636- 1636. doi: 10.3390/en11071636
18	GREY V H , CARLOS M , GONZALO N . A review on the long short-term memory model[J]. Artifitial Intellegence Review, 2020, 53 (8): 5929- 5955. doi: 10.1007/s10462-020-09838-1
19	熊明兰, 王华伟, 徐怡, 等. 基于鸟击事故征候预测的通用航空安全研究[J]. 系统工程与电子技术, 2020, 42 (9): 2033- 2040.
	XIONG M L , WANG H W , XU Y , et al. General aviation safety research based on prediction of bird strike symptom[J]. Systems Engineering and Electronics, 2020, 42 (9): 2033- 2040.
20	HOCHREITER S , SCHMIDHUBER J . Long short-term memory[J]. Nerual Computation, 1997, 9 (8): 1735- 1780. doi: 10.1162/neco.1997.9.8.1735
21	YONG Z Z . Long short-term memory recurrent neural network for remaining useful life prediction of lithium-ion batteries[J]. IEEE Trans.on Vehicular Technology, 2018, 67 (7): 5695- 5705. doi: 10.1109/TVT.2018.2805189
22	FISCHER T , KRAUSS C . Deep learning with long short-term memory networks for financial market predictions[J]. European Journal of Operational Research, 2018, 270 (2): 654- 669. doi: 10.1016/j.ejor.2017.11.054
23	LIU H , MI X W , LI Y F . Smart multi-step deep learning model for wind speed forecasting based on variational mode decomposition, singular spectrum analysis, LSTM network and ELM[J]. Energy Conversion and Management, 2018, 159 (3): 54- 64.
24	ZHAO J C , DENG F , CAI Y Y , et al. Long short-term memory-fully connected (LSTM-FC) neural network for pm2.5 concentration prediction[J]. Chemosphere, 2019, 220 (4): 486- 492.
25	ZHOU Y F , ZHANG M C , ZHU J L , et al. A randomized block-coordinate Adam online learning optimization algorithm[J]. Neural Computing & Applications, 2020, 32 (16): 12671- 12684.
26	单磊. 深度学习算法并行优化技术及应用研究[D]. 长沙: 国防科技大学, 2019: 17-18.
	SHAN L. Research on parallel optimizaion for deep learning algorithms and applications[D]. Changsha: National University of Defense Technology, 2019: 17-18.
27	薛宇敬阳. 我国通用航空飞行事故原因研究[D]. 北京: 中国矿业大学, 2019: 36-37.
	XUEYU J Y. The study on causes of general aviation flight accidents in china[D]. Beijing: China University of Mining & Technology, 2019: 36-37.
28	GREFF K , SRIVASTAVA R K , KOUTNIK J . LSTM: a search space odyssey[J]. IEEE Trans.on Neural Networks and Learning Systems, 2017, 28 (10): 2222- 2232. doi: 10.1109/TNNLS.2016.2582924
29	GAO L L , GUO Z , ZHANG H W , et al. Video captioning with attention-based LSTM and semantic consistency[J]. IEEE Trans.on Multimedia, 2017, 19 (9): 2045- 2055. doi: 10.1109/TMM.2017.2729019
30	SRIVASTAVA N , HINTON G , KRIZHEVSKY A , et al. Dropout: a simple way to prevent neural networks from overfitting[J]. Journal of Machine Learing Research, 2016, 15 (6): 1929- 1958.
31	HUANG G B . An insight into extreme learning machines: Random neurons, random features and kernels[J]. Cognitive Computation, 2014, 6 (3): 376- 390. doi: 10.1007/s12559-014-9255-2

周次	外来影响因素	设备设施因素	环境因素	管理因素	人为因素	强制报告事件
1	0.192 3	0.000 0	0.000 0	0.000 0	0.119 0	0.000 0
2	0.333 3	0.182 9	0.333 3	0.232 1	0.166 7	0.442 0
3	0.384 6	0.148 6	0.058 8	0.178 6	0.285 7	0.507 2
⋮	⋮	⋮	⋮	⋮	⋮	⋮
47	0.168 7	0.605 1	0.196 1	0.142 9	0.309 5	0.289 9
48	0.204 8	0.651 3	0.156 9	0.250 0	0.261 9	0.644 9

致因事件	相关性评价指标
致因事件	Pearson	Spearman	Kendall
外来影响因素	0.836	0.853	0.751
设备设施因素	0.728	0.706	0.478
环境因素	0.189	0.193	0.157
管理因素	0.124	0.135	0.108
人为因素	0.875	0.906	0.792

常变量	隐含节点数	批尺寸
常变量	隐含节点数	1	2	3	…	9	10
隐层数：1 步长：4 训练次数：200	3	9.852 0	9.167 6	9.003 3	…	8.458 0	8.424 9
	4	7.876 2	6.832 8	6.757 9	…	6.653 4	6.690 6
	5	9.052 8	7.073 2	5.904 0	…	3.326 4	3.028 5
	⋮	⋮	⋮	⋮	⋱	⋮	⋮
	11^*	3.878 2	1.661 3	1.535 7	…	1.476 9	1.541 8
	12	4.699 6	2.464 9	1.459 5	…	1.456 7	1.542 7

常变量	隐层数	批尺寸
常变量	隐层数	1	2	3	…	9	10
步长：4 训练次数：200	1	3.878 2	1.661 3	1.535 7	…	1.476 9	1.541 8
	2	4.027 5	1.481 8	1.271 1	…	1.463 7	1.501 0
	3	5.177 8	1.229 2^**	1.315 2	…	1.355 3	1.508 9
	⋮	⋮	⋮	⋮	⋱	⋮	⋮
	9	13.058 9	11.656 4	11.201 2	…	9.850 7	9.730 1
	10	13.893 8	12.335 2	9.787 1	…	6.212 8	6.144 7

模型	测试样本点
模型	样本点1	样本点2	样本点3	样本点4
ML-LSTM	0.028 4	0.145 4	0.031 2	0.072 1
GRU	0.143 6	0.229 5	0.005 3	0.143 6
RNN	0.143 4	0.119 0	0.094 3	0.143 4
LSTM	0.059 3	0.213 9	0.054 2	0.047 5
BP	0.010 9	0.405 0	0.248 7	0.261 4
RBF	0.110 3	0.114 7	0.350 9	0.073 2
ARIMA	0.048 2	0.243 0	0.197 3	0.086 8