关于「西格蒙德近似与洛登下界:检测延迟只随误报间隔的对数增长」,《累积和告警》给出的定量结论是什么?

结论:每个观测的信息量是 0.125。误报间隔每扩大十倍,累积和的延迟依次增加 16.2、18.1、18.4 天,逐渐逼近下界的斜率,即十的自然对数除以信息量,等于 18.4 天。固定阈值则完全是另一种增长方式:误报间隔每扩大十倍,它的延迟乘以 7.1、7.5、7.7 倍。误报间隔为十万天时,累积和平均 67.5 天报警,固定阈值要 12003.2 天。

数据口径:书中参数为示意;全部计算由书中程序生成;本书属于非个人化经营建议,第八章只做离线模型,不访问任何服务器。 书中以标注框区分已发表的证据、可运行的 Python 计算(23 段程序)、原典与作者的假说。

English version

对照数据表

下表数字由本章 Python 程序原样打印,可重新运行复现。 程序只用 Python 标准库与结论吻合的打印值: 0.125, 12003.2, 16.2, 18.1, 18.4, 67.5

ARL0 h CUSUM ARL1 fixed ARL1 log(ARL0)/I
1004.415 14.8 29.5 36.8
10008.582 31.1 208.5 55.3
1000013.109 49.1 1554.8 73.7
10000017.703 67.5 12003.2 92.1
information per reading I0.125
CUSUM delay added per tenfold ARL016.2, 18.1, 18.4
slope of the bound, ln(10) / I18.4
fixed threshold delay ratio per tenfold ARL07.1, 7.5, 7.7
查看完整验算程序
# Chapter 4. p43: delay grows with the log of the false-alarm interval.
# Model: for a shift of delta = 0.5 the CUSUM uses k = delta / 2 = 0.25,
# h from Siegmund's formula (p41) for each ARL0; its ARL1 is then taken
# from the Brook and Evans chain (m = 80). The fixed threshold uses L
# with P(Z > L) = 1 / ARL0. Lorden's (1971) asymptote for the best
# possible delay is log(ARL0) / I, with I = delta^2 / 2 the information
# per reading (illustrative).
import math


def cdf(x):
    return 0.5 * math.erfc(-x / math.sqrt(2.0))


def upper(x):
    return 0.5 * math.erfc(x / math.sqrt(2.0))


def solve(a, b):
    # Gaussian elimination with partial pivoting; back substitution
    # adds its terms with math.fsum
    n = len(b)
    m = [list(row) + [b[i]] for i, row in enumerate(a)]
    for p in range(n):
        q = max(range(p, n), key=lambda r: abs(m[r][p]))
        m[p], m[q] = m[q], m[p]
        piv = m[p]
        for r in range(p + 1, n):
            f = m[r][p] / piv[p]
            if f != 0.0:
                m[r] = m[r][:p] + [x - f * y for x, y in
                                   zip(m[r][p:], piv[p:])]
    x = [0.0] * n
    for i in range(n - 1, -1, -1):
        s = math.fsum(m[i][j] * x[j] for j in range(i + 1, n))
        x[i] = (m[i][n] - s) / m[i][i]
    return x


def arl(k, h, delta, m=80):
    w = 2.0 * h / (2 * m - 1)
    a = []
    for i in range(m):
        c = i * w - k + delta
        row = [-cdf(w / 2.0 - c)]
        for j in range(1, m):
            row.append(cdf(j * w - w / 2.0 - c)
                       - cdf(j * w + w / 2.0 - c))
        row[i] = row[i] + 1.0
        a.append(row)
    return solve(a, [1.0] * m)[0]


def siegmund(k, h, delta):
    # Siegmund (1985): b = h + 1.166, D = delta - k
    b = h + 1.166
    d = delta - k
    if abs(d) < 1e-12:
        return b * b
    return (math.exp(-2.0 * d * b) + 2.0 * d * b - 1.0) / (2.0 * d * d)


def limit_for(arl0):
    lo, hi = 0.0, 10.0
    for _ in range(200):
        mid = 0.5 * (lo + hi)
        if upper(mid) > 1.0 / arl0:
            lo = mid
        else:
            hi = mid
    return 0.5 * (lo + hi)


def h_siegmund(k, target):
    lo, hi = 0.0, 60.0
    for _ in range(100):
        mid = 0.5 * (lo + hi)
        if siegmund(k, mid, 0.0) < target:
            lo = mid
        else:
            hi = mid
    return 0.5 * (lo + hi)


if __name__ == "__main__":
    d = 0.5
    k = d / 2.0
    info = d * d / 2.0
    print("  ARL0     h    CUSUM ARL1  fixed ARL1  log(ARL0)/I")
    cs = []
    for a0 in (100.0, 1000.0, 10000.0, 100000.0):
        h = h_siegmund(k, a0)
        c = arl(k, h, d)
        f = 1.0 / upper(limit_for(a0) - d)
        cs.append((c, f))
        print(" {:6.0f}  {:6.3f}  {:9.1f}  {:10.1f}  {:10.1f}".format(
            a0, h, c, f, math.log(a0) / info))
    print("information per reading I = {:.3f}".format(info))
    steps = [b[0] - a[0] for a, b in zip(cs, cs[1:])]
    print("CUSUM delay added per tenfold ARL0: " + ", ".join(
        "{:.1f}".format(x) for x in steps))
    print("slope of the bound, ln(10) / I = {:.1f}".format(
        math.log(10.0) / info))
    print("fixed threshold delay ratio per tenfold ARL0: " + ", ".join(
        "{:.1f}".format(b[1] / a[1]) for a, b in zip(cs, cs[1:])))

唯一数据源与引用

范雅福(Fan, Y.). (2026). 《累积和控制图与平均运行长度、延迟漂移与转化暗降的最快无误报告警母典》第 4 章「西格蒙德近似与洛登下界:检测延迟只随误报间隔的对数增长」(版本 1.0.0). 5D Library. DOI 待发布

DOI 待发布 · 专著页正式发布后上线(本站暂无专著页) · 5DL-B-305 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026305,
  author    = {Fan, Yafu},
  title     = {CUSUM Control Charts and Average Run Length: the Fastest Alarm within a False-Alarm Budget for Latency Drift and Silent Conversion Drops},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

本书其他章节结论