关于「最优性的边界:自相关、残差图与大幅跳变」,《累积和告警》给出的定量结论是什么?
结论:组合规则的累积和告警限从 4.096 上调到 4.148,误报间隔仍为 370.0 天。代价落在小漂移上,而且很小:半个标准差漂移时,组合规则 28.0 天,纯累积和 27.7 天。收益落在大跳变上:漂移四个标准差时,组合规则 1.3 天,纯累积和 1.7 天,固定阈值 1.1 天;漂移五个标准差时,三者分别是 1.1、1.3、1.0 天。组合规则在全表中从未是最慢的那一个,也几乎从未明显慢于最快的那一个。
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。
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: 1.0, 1.1, 1.3, 1.7, 27.7, 28.0
| CUSUM h = 4.096; combined h = 4.148 with cap 3.5 | |||
| combined ARL0 = 370.0 | |||
| shift | fixed | CUSUM | combined |
| 0.50 | 88.9 | 27.7 | 28.0 |
| 1.00 | 26.7 | 8.6 | 8.6 |
| 2.00 | 4.6 | 3.4 | 3.3 |
| 3.00 | 1.7 | 2.2 | 2.0 |
| 4.00 | 1.1 | 1.7 | 1.3 |
| 5.00 | 1.0 | 1.3 | 1.1 |
查看完整验算程序
# Chapter 6. p63: big jumps favour the fixed threshold, so combine them.
# Model: z ~ N(delta, 1). Three rules share ARL0 = 370: the fixed
# threshold (L = 2.782, solved here), the CUSUM of p12 (k = 0.5,
# h = 4.096), and a combined rule that alarms when z > 3.5 or when the
# CUSUM passes its own re-solved limit. For the combined rule the Brook
# and Evans chain keeps only the moves with z <= 3.5, so a single large
# reading ends the run (m = 60, 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, cap=None, m=60):
# cap: also alarm at once when the reading z exceeds cap
w = 2.0 * h / (2 * m - 1)
top = 50.0 if cap is None else cap - delta
a = []
for i in range(m):
c = i * w - k + delta
row = [-cdf(min(w / 2.0 - c, top))]
for j in range(1, m):
lo = j * w - w / 2.0 - c
hi = min(j * w + w / 2.0 - c, top)
row.append(-max(0.0, cdf(hi) - cdf(lo)))
row[i] = row[i] + 1.0
a.append(row)
return solve(a, [1.0] * m)[0]
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_for(k, target, cap):
lo, hi = 0.5, 15.0
for _ in range(30):
mid = 0.5 * (lo + hi)
if arl(k, mid, 0.0, cap) < target:
lo = mid
else:
hi = mid
return 0.5 * (lo + hi)
if __name__ == "__main__":
L, k = limit_for(370.0), 0.5
h1 = h_for(k, 370.0, None)
h2 = h_for(k, 370.0, 3.5)
print("CUSUM h = {:.3f}; combined h = {:.3f} with cap 3.5".format(h1, h2))
print("combined ARL0 = {:.1f}".format(arl(k, h2, 0.0, 3.5)))
print(" shift fixed CUSUM combined")
for d in (0.5, 1.0, 2.0, 3.0, 4.0, 5.0):
print(" {:5.2f} {:6.1f} {:6.1f} {:8.1f}".format(
d, 1.0 / upper(L - d), arl(k, h1, d), arl(k, h2, d, 3.5)))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 6 章「最优性的边界:自相关、残差图与大幅跳变」(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}
}