Evidence level: C
在宣称\”残余相位漂移存在(或不存在)\”之前,必须先回答一个前置问题:在当前序列长度与噪声水平下,这套流水线到底能检出多大的漂移?
正文
判断一个节律信号里是否残留相位漂移,和用体温计判断有没有低烧是同一个逻辑:你得先知道这只体温计的刻度能细到多少。如果它只能读出 0.1°C 的变化,那 0.05°C 的波动它读不出来——你看到\”无变化\”,不代表真的没变化。
具体到本项目的四道工序流水线(EEMD 去趋势去慢速 → 相位提取 → 替代数据检验 → 效应量估计),问题被拆成三层。第一层是注入实验:取蒙特卡洛合成的节律序列,人为注入已知大小的相位漂移(0° 到 90°,步进覆盖弱漂移区),同时扫描序列长度(短程 7 天 vs 中长程 30 天以上)和噪声水平三个轴。第二层是检出率矩阵:每一格跑足够多次,统计\”流水线报出显著漂移\”的频率,得到漂移量×序列长度×噪声的检出率(功效)曲面。第三层是等值线:在曲面上画出\”检出率≥80%\”的最小可检出漂移(MDD)等值线,做成查询表——这就是这套流水线在给定条件下的\”体温计刻度\”。
这套思路不是自创,而是时间序列方法的通行验收标准。替代数据检验的功效本身就需要蒙特卡洛评估:Moulder 等的系统比较表明,检验流程在有限样本下的表现(控 I 类错误、检出能力)必须靠模拟来标定,解析近似在小样本和非正态下不可靠(PMID 28898401)。相位估计的噪声底线也有实测参照:穿戴设备驱动的数学模型预测昼夜相位,普通作息人群误差可到 1 小时以内,但轮班人群等高节律紊乱条件下误差显著放大(PMID 34013347);更精细的多模态模型能把误差压到约 30 分钟,代价是传感器负荷上升(PMID 25570442)。把\”小时级\”换算到 24 小时周期的角度坐标,1 小时约等于 15°——这意味着在典型噪声下,小于几度的残余漂移很可能被估计误差淹没,这正是需要 MDD 查询表的根本原因。
对易学\”子午流注\”\”气机节律\”一类命题,这个标定直接决定解读边界:若观测数据的漂移量估计落在 MDD 等值线之下,正确的表述是\”低于本流水线检出能力\”,而不是\”无漂移\”;只有漂移量明确越过等值线,\”真实漂移\”的宣称才站得住。
证据与局限
已验证:替代数据类检验的功效依赖蒙特卡洛标定是统计学共识(PMID 28898401);昼夜相位估计存在明确的方法与噪声依赖误差底线(PMID 34013347, 25570442)。
部分验证:相位估计误差的量级(约 30 分钟至 1 小时)来自穿戴与生理信号场景,与脉象/气感序列的噪声结构不完全同构,直接换算需谨慎。
待验证:本项目四道工序流水线在 0-90° 注入漂移下的具体 MDD 等值线尚未实测;EEMD 去趋势环节本身是否会削掉一部分真实漂移(\”去趋势误伤\”)也需在注入实验中同步量化。该主题尚未进入 L1 对应网络。
相关阅读
证据等级: C
参考文献: PMID 28898401, PMID 34013347, PMID 25570442
证据等级:A=有PubMed直接支撑,B=科学类比/间接证据,C=传统文化共识/初步探索
L1 网络状态:经查 CORRESPONDENCE_NETWORK.md 与 EDGE_TABLE_science-yixue.md,无「残余相位漂移/EEMD 检出阈值」对应边——该主题尚未进入 L1 对应网络,证据等级封顶 C。
免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。
Detecting the Bottom Line of Residual Phase Drift: First Calibrate \”How Much Drift Counts as Real,\” Then Discuss Whether There Is Any Drift
Evidence level: C
Before claiming that “residual phase drift exists (or does not exist),” one must first answer a prerequisite question: given the current sequence length and noise level, what is the maximum drift this pipeline can actually detect?
Main Body
Determining whether a rhythm signal contains residual phase drift follows the same logic as using a thermometer to check for a low-grade fever: you must first know how fine the scale of that thermometer is. If it can only read changes as small as 0.1°C, then a fluctuation of 0.05°C will go undetected—observing “no change” does not mean there is truly no change.
Specifically, for this project’s four-step pipeline (EEMD detrending/removal of slow components → phase extraction → surrogate data testing → effect size estimation), the problem is decomposed into three layers. The first layer is the injection experiment: Monte Carlo-synthesized rhythm sequences are used, with known phase drifts artificially injected (ranging from 0° to 90°, covering the weak-drift region in steps), while simultaneously scanning across three axes: sequence length (short-term 7 days vs. medium-to-long-term 30+ days) and noise level. The second layer is the detection rate matrix: each cell is run sufficient times to statistically determine the frequency of the pipeline reporting “significant drift,” yielding a detection rate (power) surface as a function of drift magnitude × sequence length × noise. The third layer is the contour lines: on this surface, contours of “detection rate ≥ 80%” are drawn to define the Minimum Detectable Drift (MDD), creating a lookup table—this constitutes the “thermometer scale” of this pipeline under given conditions.
This approach is not original but aligns with standard acceptance criteria for time-series methods. The power of surrogate data testing itself requires Monte Carlo calibration: systematic comparisons by Moulder et al. show that performance under finite samples (controlling Type I error, detection power) must be calibrated via simulation, as analytical approximations are unreliable for small samples and non-Gaussian distributions (PMID 28898401). Noise floors for phase estimation also have empirical references: mathematical models driven by wearable devices predict circadian phase, with errors reaching within 1 hour for typical populations but significantly amplifying under high-rhythm-disorder conditions such as shift workers (PMID 3403347); more refined multimodal models can reduce error to approximately 30 minutes at the cost of increased sensor load (PMID 25570442). Converting “hour-level” error to the angular coordinate of a 24-hour cycle, 1 hour corresponds roughly to 15°—meaning that under typical noise conditions, residual drifts smaller than a few degrees are likely to be submerged by estimation error, which is precisely why an MDD lookup table is essential.
For Yijing Studies propositions such as “Ziwu Liuzhu” (a traditional pattern of energy flow tied to time and direction) or “Qiji Jielu” (rhythmic patterns of vital energy), this calibration directly determines the boundaries of interpretation: if the estimated drift in observational data falls below the MDD contour, the correct statement is “below this pipeline’s detection capability,” rather than “no drift”; only when the drift magnitude clearly exceeds the contour can claims of “real drift” stand.
Evidence and Limitations
Verified: The reliance of surrogate data testing power on Monte Carlo calibration is a statistical consensus (PMID 28898401); clear methodological and noise-dependent error floors exist for circadian phase estimation (PMID 3403347, PMID 25570442).
Partially Verified: The magnitude of phase estimation error (approximately 30 minutes to 1 hour) originates from wearable and physiological signal contexts, which do not fully match the noise structure of pulse waveforms or Qi-sensation sequences; direct conversion requires caution.
To Be Verified: The specific MDD contours for this project’s four-step pipeline under 0–90° injected drift have not yet been empirically measured; whether the EEMD detrending step itself suppresses some true drift (“detrending collateral damage”) also needs simultaneous quantification in injection experiments. This topic has not yet entered the L1 correspondence network.
Further Reading
- How Training Duration and Dosage Calibrate Effort Asymmetry
- Independence Test: Repeated Practice Outpaces Initial Learning, Recovery Slower Than Restoration
- Research Pool
Evidence Level: C
References: PMID 28898401, PMID 3403347, PMID 25570442
Evidence Level: A = direct PubMed support, B = scientific analogy/indirect evidence, C = traditional consensus/preliminary exploration
L1 Network Status: Checked CORRESPONDENCE_NETWORK.md and EDGE_TABLE_science-yixue.md; no edge corresponding to “residual phase drift/EEMD detection threshold” exists—this topic has not yet entered the L1 correspondence network, capping evidence level at C.
Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.
Leave a Reply