关于「会放弃的队列:Erlang C、Erlang-A 与平方根配员」,《私域等待、转人工门限与客诉阻断》给出的定量结论是什么?
结论:第二章用 Erlang-A 说明,同一支六人团队,Erlang C 给出 103.6 秒的平均等待,计入放弃后只有 13.1 秒,代价是 0.109 的高意向买手在等待中离开。
数据口径:书中参数为示意;全部计算由书中程序生成;本书不是法律、财务或经营上的个人化建议。 书中以标注框区分已发表的证据、可运行的 Python 计算(18 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: 0.109, 103.6, 13.1
| load 4.8 erlangs, mean patience 120 s | ||||
| agents | P(wait) | P(abandon) | mean wait (s) | P(ab)/E[W] per min |
| 3 | 0.773 | 0.441 | 52.9 | 0.500 |
| 4 | 0.621 | 0.299 | 35.9 | 0.500 |
| 5 | 0.457 | 0.188 | 22.6 | 0.500 |
| 6 | 0.307 | 0.109 | 13.1 | 0.500 |
| 7 | 0.186 | 0.058 | 7.0 | 0.500 |
| 8 | 0.103 | 0.029 | 3.4 | 0.500 |
| 9 | 0.051 | 0.013 | 1.5 | 0.500 |
| 6 agents: 0.109 of high-intent buyers leave unanswered; | ||||
| Erlang C (p21) shows a 103.6 s wait for the same desk, Erlang-A 13.1 s |
查看完整验算程序
# Chapter 2. p22: Erlang-A (M/M/n+M, the Palm model) on the same desk.
# Model (illustrative parameters): as in p21, plus exponential patience
# with mean PAT minutes: a waiting buyer leaves at rate THETA = 1/PAT.
# The number in the system is a birth-death chain: births at LAM,
# deaths at min(k, n) MU + max(k - n, 0) THETA. The chain is solved
# exactly on 0..K (K large enough that the tail is below 1e-15).
# Little's law gives the mean wait; abandonment = THETA E[queue] / LAM,
# so P(abandon) / E[wait] equals THETA exactly in this model.
import math
LAM, HANDLE, PAT, K = 1.2, 4.0, 2.0, 400
MU, THETA = 1.0 / HANDLE, 1.0 / PAT
def solve(n):
w = [1.0]
for k in range(1, K + 1):
death = min(k, n) * MU + max(k - n, 0) * THETA
w.append(w[-1] * LAM / death)
z = math.fsum(w)
p = [x / z for x in w]
p_wait = math.fsum(p[n:])
eq = math.fsum((k - n) * p[k] for k in range(n + 1, K + 1))
return p_wait, THETA * eq / LAM, eq / LAM
print("load {:.1f} erlangs, mean patience {:.0f} s".format(
LAM * HANDLE, 60.0 * PAT))
print(" agents P(wait) P(abandon) mean wait (s) P(ab)/E[W] per min")
for n in range(3, 10):
pw, pab, ew = solve(n)
print("{:7d} {:7.3f} {:10.3f} {:13.1f} {:18.3f}".format(
n, pw, pab, 60.0 * ew, pab / ew))
pw, pab, ew = solve(6)
print("6 agents: {:.3f} of high-intent buyers leave unanswered;".format(
pab))
print("Erlang C (p21) shows a 103.6 s wait for the same desk, Erlang-A"
" {:.1f} s".format(60.0 * ew))唯一数据源与引用
范雅福(Fan, Y.). (2026). 《私域对话队列耐心衰减分布、人机协同红黄绿灯移交门限与客诉升级马尔可夫阻断》第 2 章「会放弃的队列:Erlang C、Erlang-A 与平方根配员」(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-175 · v1.0.0 · CC-BY-NC-ND-4.0
BibTeX
@book{fan2026175,
author = {Fan, Yafu},
title = {Patience-Decay Distributions in Private-Message Queues, Red-Amber-Green Human-AI Handoff Thresholds and Markov Blocking of Complaint Escalation},
year = {2026},
publisher = {5D Library},
edition = {1.0.0},
note = {DOI pending}
}