数据驱动的间歇低氧训练贝叶斯优化决策方法

陈婧; 史大威; 蔡德恒; 王军政; 朱玲玲

doi:10.16383/j.aas.c220712

数据驱动的间歇低氧训练贝叶斯优化决策方法

doi: 10.16383/j.aas.c220712

1.
北京理工大学自动化学院工业和信息化部伺服运动系统驱动与控制重点实验室北京 100081
2.
中国人民解放军军事科学院军事医学研究院北京 100850

基金项目: 国家自然科学基金(61973030), 北京市科技计划项目 (Z161100000216134)资助

详细信息

作者简介:
陈婧：北京理工大学自动化学院博士研究生. 主要研究方向为医学信号处理和性能评估. E-mail: jingchen@bit.edu.cn

史大威：北京理工大学自动化学院教授. 主要研究方向为复杂采样控制系统分析与设计及在生物医学、机器人及运动系统中的应用. 本文通信作者. E-mail: daweishi@bit.edu.cn

蔡德恒：北京理工大学自动化学院博士研究生. 主要研究方向为事件触发的采样控制、估计与学习以及闭环给药系统控制算法设计与实现. E-mail: dehengcai@bit.edu.cn

王军政：北京理工大学自动化学院教授. 主要研究方向为运动驱动与控制, 电液伺服/比例控制, 试验测试与负载模拟, 机器人控制. E-mail: wangjz@bit.edu.cn

朱玲玲：中国人民解放军军事科学院军事医学研究院研究员. 主要研究方向为高原等特殊环境对机体损伤与防护措施的研究. E-mail: zhull@bmi.ac.cn

计量
- 文章访问数: 482
- HTML全文浏览量: 122
- PDF下载量: 151
- 被引次数: 0
出版历程
- 收稿日期: 2022-09-08
- 录用日期: 2023-02-10
- 网络出版日期: 2023-03-21
- 刊出日期: 2023-08-21

Data-driven Bayesian Optimization Method for Intermittent hypoxic Training Strategy Decision

1.
Ministry of Industry and Information Technology Key Laboratory of Servo Motion System Drive and Control, School of Automation, Beijing Institute of Technology, Beijing 100081
2.
Academy of Military Medical Sciences, Beijing 100850

Funds: Supported by National Natural Science Foundation of China (61973030) and Beijing Municipal Science and Technology Commission (Z161100000216134)

More Information

Author Bio:
CHEN Jing　Ph.D. candidate at the School of Automation, Beijing Institute of Technology. Her research interest covers medical signal processing and performance assessment

SHI Da-Wei　Professor at the School of Automation, Beijing Institute of Technology. His research interest covers analysis & design of advanced sampled-data control systems, with applications to biomedical engineering, robotics and motion systems. Corresponding author of this paper

CAI De-Heng　Ph.D. candidate at the School of Automation, Beijing Institute of Technology. His research interest covers event-triggered sampled-data control, state estimation and machine learning, and the control algorithm design and implementation of closed-loop drug delivery systems

WANG Jun-Zheng　Professor at the School of Automation, Beijing Institute of Technology. His research interest covers motion drive and control, electro-hydraulic servo/proportional control, test experiment and load simulation, and robotic control

ZHU Ling-Ling　Professor of Academy of Military Medical Sciences. Her research interest covers body damage and protective measures in high-altitude environment

摘要

摘要: 青藏地区快速的经济发展使得进入高原的群体数量日益增加, 随之而来的高原健康问题也愈发突出. 间歇性低氧训练(Intermittent hypoxic training, IHT)是急进高原前常使用的预习服方法, 一般针对不同个体均设置固定的开环策略, 存在方案制定无标准、系统化的理论指导缺乏、效果不明显等问题. 针对以上情况, 设计了一种小样本数据驱动的IHT策略贝叶斯闭环学习优化框架, 建立自回归结构的高斯过程血氧饱和度(Peripheral oxygen saturation, SpO₂)预测模型, 并考虑高低风险事件对训练的影响, 设计与氧浓度变化方向和速率相关的风险不对称代价函数, 提出具有安全约束的贝叶斯优化方法, 实现IHT最优供氧浓度的优化决策. 考虑到现有仿真器无法反映个体动态变化过程, 依据“最优速率理论”设计了合理的模型自适应变化律. 所提出闭环干预方法通过该仿真器进行了可行性和有效性验证. 说明该学习框架能够指导个体提升高原适应能力, 减轻其在预习服阶段的非适应性不良反应, 为个性化IHT提供精准调控手段.
- 数据驱动控制 /
- 高斯过程 /
- 贝叶斯优化 /
- 风险不对称代价函数 /
- 高原适应能力提升 /
- 间歇性低氧训练
Abstract: The rapid economic development of Qinghai-Tibet region has led to an increasing number of groups entering the plateau, and the consequent problem of high-altitude health has become increasingly prominent. Intermittent hypoxic training (IHT) is a commonly-used preacclimatization approach before rapidly going to the plateau. It is usually designed as fixed open-loop strategies for different individuals, which has several disadvantages such as no standard formulation, lack of systematic theoretical guidance and poor efficacy. In this paper, a data-driven Bayesian closed-loop learning optimization framework of IHT strategy is designed by using small samples, and a Gaussian process model with autoregressive structure of peripheral oxygen saturation (SpO₂) is built for prediction. Based on the predictive model, a risk asymmetric cost function related to the oxygen concentration rate and its direction is developed. Finally, a Bayesian optimization method with safety constraints is proposed to enable the optimal decision of IHT oxygen concentration. Given that the existing simulator cannot reflect the process dynamics of individuals, a reasonable model adaptation law is designed according to the “optimal rate theory”. The feasibility and effectiveness of the proposed closed-loop intervention method are verified by the simulator. These results indicate that the proposed learning framework can help individuals to improve their adaptability to high-altitudes, reduce their non-adaptive adverse reactions in the pretraining stage, and provide precise control solution to personalized IHT.
- Data-driven control /
- Gaussian process /
- Bayesian optimization /
- risk asymmetric cost function /
- high-altitude adaptability improvement /
- intermittent hypoxic training (IHT)

HTML全文

图 1 高原适应性能力提升的IHT策略优化决策算法流程图

Fig. 1 Flow chart of IHT optimization decision algorithm for high-altitude adaptability improvement

下载: 全尺寸图片幻灯片

图 2 高斯过程预测算法示意图

Fig. 2 Flow chart of Gaussian process prediction algorithm

下载: 全尺寸图片幻灯片

图 3 所设计代价函数$ {\cal{L}}_{v} $部分的惩罚强度在不同$ \Delta c $下的变化

Fig. 3 The penalty changes of designed $ {\cal{L}}_v $ term under different $ \Delta c $ values

下载: 全尺寸图片幻灯片

图 4 虚拟受试者1和2采取开环和闭环策略进行IHT的$ {\rm{ SpO}}_2 $曲线

Fig. 4 The $ {\rm{ SpO}}_2 $ curves of simulated subject 1 and 2 that perform IHT based on traditional open-loop strategy and proposed closed-loop strategy

下载: 全尺寸图片幻灯片

图 5 虚拟受试者3和4采取开环和闭环策略进行IHT的$ {\rm{ SpO}}_2 $曲线

Fig. 5 The $ {\rm{ SpO}}_2 $ curves of simulated subject 3 and 4 that perform IHT based on traditional open-loop strategy and proposed closed-loop strategy

下载: 全尺寸图片幻灯片

图 6 10名虚拟受试者采取开环和闭环策略进行IHT的$ {\rm{ SpO}}_2 $曲线

Fig. 6 The $ {\rm{ SpO}}_2 $ curves of 10 simulated subjects that perform IHT based on traditional open-loop strategy and proposed closed-loop strategy

下载: 全尺寸图片幻灯片

表 1 相关参数取值

Table 1 Related parameters

参数	含义	取值
$ n_s $	一日IHT的低氧段总数	8
$ n_t $	一段低氧段预测点总数	200
$ n_p $	一段低氧段采样点总数	300
$ \sigma^{2} $	平方指数核函数超参数	10
$ \ell $	平方指数核函数超参数	10
$ \boldsymbol{y}_{r} $	目标$ {\rm{ SpO}}_2 $向量	$ [95H_{50},90H_{50}, 85H_{100}]^\mathrm{T} $
$ Q $	代价函数惩罚矩阵	136I
$ w $	代价函数权重系数	1000
$ c_{l,\min} $	供氧浓度下阈值 (%)	10
$ c_{l,\max} $	供氧浓度上阈值 (%)	15

下载: 导出CSV

表 2 虚拟受试者1和2采取开环策略进行IHT和采取闭环策略进行IHT的效果对比

Table 2 Comparison results of simulated subject 1 and 2 trained by using traditional open-loop strategy and proposed closed-loop strategy

个体设定	指标	初始状态	开环策略	闭环策略
$ c_{o}=13 $, $ \Delta c_{op}=-1.5 $	DSI (s)	137	84	256
	SpO₂平均值(%)	90.9	84.9	93.6
	SpO₂标准差(%)	5.5	5.4	4.8
$ c_{o}=12 $, $ \Delta c_{op}=-1.5 $	DSI (s)	137	90	300
	SpO₂平均值(%)	90.9	87.8	96.2
	SpO₂标准差(%)	5.5	5.4	3.6

下载: 导出CSV

表 3 虚拟受试者3和4采取开环策略进行IHT和采取闭环策略进行IHT的效果对比

Table 3 Comparison results of simulated subject 3 and 4 trained by using traditional open-loop strategy and proposed closed-loop strategy

个体设定	指标	初始状态	开环策略	闭环策略
$ c_{o}=12 $, $ \Delta c_{op}=-1 $	DSI (s)	154	136	203
	SpO₂平均值(%)	91.4	89.5	94.7
	SpO₂标准差(%)	6.1	6.5	5.0
$ c_{o}=12 $, $ \Delta c_{op}=-1.5 $	DSI (s)	154	81	300
	SpO₂平均值 (%)	91.4	89.5	96.4
	SpO₂标准差(%)	6.1	6.6	4.1

下载: 导出CSV

表 4 10名虚拟受试者采取开环策略进行IHT和采取闭环策略进行IHT的效果对比

Table 4 Comparison results of 10 simulated subjects trained by using traditional open-loop strategy and proposed closed-loop strategy

受试者	指标	初始状态	开环策略	闭环策略
1	DSI (s)	137	32	300
	SpO₂平均值(%)	90.9	84.9	96.4
	SpO₂标准差(%)	5.46	5.37	3.53
2	DSI (s)	140	53	300
	SpO₂平均值(%)	92.7	86.1	95.6
	SpO₂标准差(%)	4.72	4.85	3.73
3	DSI (s)	138	31	300
	SpO₂平均值(%)	92.3	84.8	97.4
	SpO₂标准差(%)	4.81	4.70	2.88
4	DSI (s)	271	153	300
	SpO₂平均值(%)	94.6	89.9	98.5
	SpO₂标准差(%)	4.04	4.10	2.29
5	DSI (s)	138	35	184
	SpO₂平均值(%)	92.3	87.1	96.3
	SpO₂标准差(%)	4.32	4.33	3.51
6	DSI (s)	138	20	279
	SpO₂平均值(%)	92.3	84.8	94.1
	SpO₂标准差(%)	4.81	4.79	4.34
7	DSI (s)	42	35	78
	SpO₂平均值(%)	89.0	85.1	91.5
	SpO₂标准差(%)	5.44	5.52	5.19
8	DSI (s)	91	84	176
	SpO₂平均值(%)	89.9	87.6	92.1
	SpO₂标准差(%)	6.17	6.16	5.63
9	DSI (s)	138	66	300
	SpO₂平均值(%)	91.3	88.2	95.1
	SpO₂标准差(%)	5.01	4.86	4.02
10	DSI (s)	138	44	300
	SpO₂平均值(%)	91.3	84.8	96.0
	SpO₂标准差(%)	4.94	4.78	3.64

下载: 导出CSV

表 5 10名虚拟受试者的对比结果

Table 5 Comparison results of 10 simulated subjects

指标	初始状态	开环策略	闭环策略
DSI (s)	137.1	56.3	251.7
SpO₂平均值(%)	91.7	86.3	95.3
SpO₂标准差(%)	4.97	4.95	3.88

下载: 导出CSV

参考文献(33)

[1]	Andrew M L, Peter H H. Medical conditions and high-altitude travel. New England Journal of Medicine, 2022, 386(4): 364-373 doi: 10.1056/NEJMra2104829
[2]	Joshua C T, Philip N A. Global and country-level estimates of human population at high altitude. Proceedings of the National Sciences, 2021, 118(18): e2102463118 doi: 10.1073/pnas.2102463118
[3]	Cobb A B, Levett D Z H, Mitchell K, Aveling W, Hurlbut D, Gilbert-Kawai E, et.al. Physiological responses during ascent to high altitude and the incidence of acute mountain sickness. Physiological reports, 2021, 9(7): e14809
[4]	Gudbjartsson T, Sigurdsson E, Gottfredsson M, Bjornsson O M, Gudmundsson G. High altitude illness and related diseases - A review. Laeknabladid, 2019, 105(11): 499-507
[5]	Victor S, Jan C P, and Katarína K. Manifestation of intracranial lesions at high altitude: Case report and review of the literature. High Altitude Medicine & Biology, 2021, 22(1): 87-89
[6]	Fulco C S, Beidleman B A, Muza S R. Effectiveness of preacclimatization strategies for highaltitude exposure. Exercise and Sport Sciences Reviews, 2013, 41(1): 55-63 doi: 10.1097/JES.0b013e31825eaa33
[7]	Ambroży T, Maciejczyk M, Klimek A T, Wiecha S, Stanula A, Snopkowski P, et.al. The effects of intermittent hypoxic training on anaerobic and aerobic power in boxers. International Journal of Environmental Research and Public Health, 2020, 17(24): 9361 doi: 10.3390/ijerph17249361
[8]	Wille M, Gatterer H, Mairer K, Philippe M, Schwarzenbacher H, Faulhaber M, et.al. Short-term intermittent hypoxia reduces the severity of acute mountain sickness. Medicine & Science in Sports, 2012, 22(5): e79-e85
[9]	刘园园. 高原健康理论框架下的渐进型间歇性低氧预习服训练研究 [博士学位论文], 山东大学, 中国, 2014 Liu Yuan-Yuan. Short-Term Intermittent Hypoxia Reduces the Severity of Acute Mountain Sickness [Ph.D. dissertation], Shandong University, China, 2014
[10]	Treml B, Kleinsasser A, Hell T, Knotzer H, Wille M, Burtscher M. Carry-over quality of pre-acclimatization to altitude elicited by intermittent hypoxia: A participant-blinded, randomized controlled trial on antedated acclimatization to altitude. Frontiers in Physiology, DOI: 10.3389/fphys.2020.00531
[11]	Gangwar A, Pooja, Sharma M, Singh K, Patyal A, Bhaumik G, et.al. Intermittent normobaric hypoxia facilitates high altitude acclimatization by curtailing hypoxia-induced infammation and dyslipidemia. Pflugers Archiv, 2019, 471(7):949-959 doi: 10.1007/s00424-019-02273-4
[12]	杨军, 俞梦孙, 曹征涛, 吴峰, 张宏金, 王海涛, 等.间歇性递增式常压低氧暴露训练对高原习服效果的研究. 中华航空航天医学杂志, 2012, 3: 161-164 Yang Jun, Yu Meng-Sun, Cao Zheng-Tao, Wu Feng, Zhang Hong-Jin, Wang Hai-Tao, et.al. Study on the effect of increasing intermittent hypoxia exposure on altitude acclimatization. Chinese Journal of Aerospace Medicine, 2012, 3: 161-164
[13]	Kwiatkowska M, Atkins M S, Ayas N T, Ryan C F. Knowledge-based data analysis: First step toward the creation of clinical prediction rules using a new typicality measure. IEEE Transactions on Information Technology in Biomedicine, 2007, 11(6):651-660 doi: 10.1109/TITB.2006.889693
[14]	Sakellarios A I, Räber L, Bourantas C V, Exarchos T P, Athanasiou L S, Pelosi G, et.al. Prediction of atherosclerotic plaque development in an In Vivo coronary arterial segment based on a multilevel modeling approach. IEEE Transactions on Biomedical Engineering, 2017, 64(8):1721-1730 doi: 10.1109/TBME.2016.2619489
[15]	喻勇, 司小胜, 胡昌华, 崔忠马, 李洪鹏. 数据驱动的可靠性评估与寿命预测研究进展:基于协变量的方法. 自动化学报, 2018, 44(2): 216-227 Yu Yong, Si Xiao-Sheng, Hu Chang-Hua, Cui Zhong-Ma, Li Hong-Peng. Data driven reliability assessment and life-time prognostics: A review on covariate models. Acta Automatica Sinica, 2018, 44(2): 216-227
[16]	李天梅, 司小胜, 刘翔, 裴洪. 大数据下数模联动的随机退化设备剩余寿命预测技术. 自动化学报, 2022, 48(9): 2119-2141 doi: 10.16383/j.aas.c201068 Li Tian-Mei, Si Xiao-Sheng, Liu Xiang, Pei Hong. Data-model interactive remaining useful life prediction technologies for stochastic degrading devices with big data. Acta Automatica Sinica, 2022, 48(9): 2119-2141 doi: 10.16383/j.aas.c201068
[17]	蒋珂, 蒋朝辉, 谢永芳, 潘冬, 桂卫华. 基于动态注意力深度迁移网络的高炉铁水硅含量在线预测方法. 自动化学报, DOI: 10.16383/j.aas.c210524"> 10.16383/j.aas.c210524 Jiang Ke, Jiang Zhao-Hui, Xie Yong-Fang, Pan Dong, Gui Wei-Hua. Online prediction method for silicon content of molten iron in blast furnace based on dynamic attention deep transfer network. Acta Automatica Sinica, DOI: 10.16383/j.aas.c210524"> 10.16383/j.aas.c210524
[18]	Box G E, Jenkins G M, Reinsel G C, Ljung G M. Time Series Analysis: Forecasting and Control. Hoboken: John Wiley & Sons, 2015.
[19]	Xie J, Wang Q. Benchmarking machine learning algorithms on blood glucose prediction for type I diabetes in comparison with classical time-series models. IEEE Transactions on Biomedical Engineering, 2020, 67(11): 3101-3124 doi: 10.1109/TBME.2020.2975959
[20]	Moniri A, Terracina D, Rodriguez-Manzano J, Strutton P H, Georgiou P. Real-time forecasting of sEMG features for trunk muscle fatigue using machine learning. IEEE Transactions on Biomedical Engineering, 2021, 68(2): 718-727 doi: 10.1109/TBME.2020.3012783
[21]	Michalis K T. Variational learning of inducing variables in sparse Gaussian processes. In: Proceedings of the 12th International Conference on Artificial Intelligence and Statistics. Florida, USA: PMLR, 2009. 567−574
[22]	Beckers T, Hirche S. Prediction with approximated gaussian process dynamical models. IEEE Transactions on Automatic Control, 2022, 68: 6460-6473
[23]	Lee S I, Mortazavi B, Hoffman H A, Lu D S, Li C, Paak B H, et.al. A prediction model for functional outcomes in spinal cord disorder patients using gaussian process regression. IEEE Transactions on Biomedical Engineering, 2016, 20(1): 91-99
[24]	Huang H, Song Y, Peng X, Ding S X, Zhong W, Du W, et.al. A sparse nonstationary trigonometric gaussian process regression and its application on nitrogen oxide prediction of the diesel engine. IEEE Transactions on Industrial Informatics, 2021, 17(12): 8367-8377 doi: 10.1109/TII.2021.3068288
[25]	史大威, 蔡德恒, 刘蔚, 王军政, 纪立农. 面向智能血糖管理的餐前胰岛素剂量贝叶斯学习优化方法. 自动化学报, DOI: 10.16383/j.aas.c210067 Shi Da-Wei, Cai De-Heng, Liu Wei, Wang Jun-Zheng, Ji Li-Nong. Bayesian learning based optimization of meal bolus dosage for intelligent glucose management. Acta Automatica Sinica, DOI: 10.16383/j.aas.c210067
[26]	金哲豪, 刘安东, 俞立. 基于GPR和深度强化学习的分层人机协作控制. 自动化学报, 2022, 48(9): 1-11 Jin Zhe-Hao, Liu An-Dong, Yu Li. Hierarchical human-robot cooperative control based on GPR and DRL. Acta Automatica Sinica, 2022, 48(9): 1-11
[27]	Rosolia U, Zhang X, Borrelli F. Data-driven predictive control for autonomous systems. Annual Review of Control, Robotics, and Autonomous Systems, 2018, 1(1): 259-286 doi: 10.1146/annurev-control-060117-105215
[28]	Yu M. Human-performance engineering at high altitude. Science Supp, 2014: 7−8
[29]	Chen J, Xiao R, Wang L, Zhu L, Shi D. Unveiling interpretable key performance indicators in hypoxic response: a system identification approach. IEEE Transactions on Industrial Electronics, 2022, 69(12): 13676-13685 doi: 10.1109/TIE.2021.3137618
[30]	Chen J, Tian Y, Zhang G, Cao Z, Zhu L, Shi D. IoT-enabled intelligent dynamic risk assessment of acute mountain sickness: The role of event-triggered signal processing. IEEE Transactions on Industrial Informatics, 2023, 19(1): 730−738
[31]	Williams C K I, Rasmussen C E. Gaussian Processes for Machine Learning. Cambridge: The MIT Press, 2006.
[32]	Hackett, Peter H. and Roach, Robert C. High-altitude llness. New England Journal of Medicine, 2001, 345(2): 107-114 doi: 10.1056/NEJM200107123450206
[33]	Levine B D, Stray-Gundersen J. Dose-response of altitude training: how much altitude is enough? Advances in Experimental Medicine and Biology, 2006, 69: 233-247