使用 Python 和人工智能进行仿真建模实践Hands-On Simulation Modeling with Python and AI

该课程专为零编程和统计背景的商业决策者设计，核心在于将 AI（如 Claude 或 ChatGPT）转化为战略分析的超级杠杆。学员将掌握“先规格、后生成”的 AI 工作流，无需亲自动手编写代码，即可指导 AI 构建出蒙特卡洛组合风险和离散事件运营流等高级模拟模型，从而对资本准备、人员排班或新产品发布等缺乏历史数据的未来场景，进行严谨的“What-if”情境预测与置信区间分析。

为了确保模型结果能够经受住高管层的质疑，课程独创了“阅读、断言、压力测试”三层验证法，教你如何有效识别并纠正 AI 生成代码中的逻辑漏洞。最终，学员不仅能熟练运用系统动力学或智能体建模等不同范式来拆解复杂的现实问题，还将学会如何将深奥的模拟结果浓缩进一张高度可视化的“一页纸报告”中，用商业语言说服怀疑的利益相关者，真正赢得战略决策的主导权。

---

Published 7/2026
Created by Dan Sullivan
MP4 | Video: h264, 1920×1080 | Audio: AAC, 44.1 KHz, 2 Ch
Level: All Levels | Genre: eLearning | Language: English | Duration: 59 Lectures ( 4h 57m ) | Size: 2 GB

在不确定性下解答业务问题：利用人工智能设计、生成和验证仿真模型

您将学到什么
⚡ 使用规范优先的 AI 工作流程：编写结构化的模型规范，使用 Claude 或 ChatGPT 生成可运行的 Python 代码，无需从头开始编写代码即可迭代
⚡ 使用您自己构建的模型，在不确定性下回答实际的业务问题，从资本储备到人员配置计划再到产品发布
⚡ 通过三层验证（读取、断言、压力测试）AI 生成的代码，以便您可以向持怀疑态度的利益相关者解释您的结果
⚡ 运行分析和机器学习无法实现的假设情景：模拟没有历史先例的未来，并使用置信区间报告结果
⚡ 构建具有相关违约、政策杠杆和经济衰退压力情景的蒙特卡罗贷款组合风险模型
⚡ 构建具有真实就诊率、高峰感知人员配置和运营经理仪表板的离散事件诊所患者流量模型
⚡ 为新问题选择合适的范式（蒙特卡罗、离散事件、系统动力学、基于代理）以及每个不确定输入的正确分布
⚡ 以单页交付物的形式传达模拟结果这使得该模型有权被采取行动。

要求
❗无需编程或统计学背景。您将阅读和验证代码，而不是编写代码，所有数学概念均采用图示讲解，无需微积分知识。❗
需要一个免费的AI聊天助手账户，例如Claude或ChatGPT，用于根据您的需求生成代码。❗需要
一个网络浏览器。实验在Udemy内置的Jupyter环境中运行，因此无需安装任何软件；最后一部分介绍了可选的本地设置。

人工智能
可以在几秒钟内编写出可运行的代码。接下来你的选择，要么会威胁到你的职业生涯，要么会成为你迄今为止掌握的最强大的工具。关键在于一项技能：准确了解需要提出的问题，以及如何在任何人采取行动之前验证答案。本课程将通过仿真建模来教授这项技能，即在不确定情况下，运用可验证的模型来解答业务问题。

一旦你仔细观察，就会发现这些问题无处不在。

✨ 银行应该预留多少资金来应对明年的贷款损失？

✨ 周一早上诊所需要多少位医生？

✨ 新品发布预算应该用于广告还是网红营销？

五年前，要解决这些问题，就得聘请专家。生成式人工智能改变了这一切：编写模拟代码过去需要一位资深工程师花费一周时间，而现在，对于一个具体明确的问题，只需一个下午就能搞定。瓶颈从打字转移到了思考，而思考始终是价值所在。本课程将助你成为这道价值线上的重要一环。

已经在从事数据分析或机器学习工作了吗？仿真可以弥补这些工具的不足。

✨ 分析和机器学习能够从过去吸取经验。当昨天的数据能够预测明天的世界时，它们就是合适的工具。

✨ 他们遇到的困难恰恰在于风险最高的地方：一个前所未有的决定，一个训练数据中没有先例的冲击，一个关于从未尝试过的政策的假设性问题。

✨ 模拟则反其道而行之。你先对系统本身的运行方式（到达、排队、默认设置、反馈回路）进行建模，然后模拟尚未发生的未来情况。

✨ 当利益相关者提出“如果需求翻倍会怎样？”或“严重的经济衰退会有多严重？”等问题时，基于历史数据训练的模型只能进行推断。而模拟则可以给出答案，并提供完整的置信区间。

这就是为什么模拟技术应该成为每位分析师和数据科学家工具箱中的必备技能，与你已经掌握的技能并驾齐驱。

一套工作流程，适用于所有模型。你将在所有构建过程中实践相同的五个步骤。

✨制定规范。用结构化的、通俗易懂的语言描述业务问题、假设及其来源、流程以及证明模型正确的检查方法。

✨生成它。将你的规范粘贴到 Claude 或 ChatGPT 中，即可获得可运行的 Python 代码。

✨运行它。在浏览器中的 Jupyter Notebook 中执行。Udemy 已提供环境，因此无需安装任何软件。

✨验证它。三层验证：阅读代码、运行断言、使用极端输入进行压力测试。

✨迭代。当结果引发新问题时，编辑规范并重新生成。

这种节奏感是这门课程教授的最具通用性的技能，无论模型是风险分析还是你要求 AI 为你构建的任何其他东西，它都同样适用。

你将构建什么？本课程涵盖四大仿真范式，其中两种深入讲解，两种概述。

✨蒙特卡洛（深度实践）：一家社区银行的贷款组合风险模型，由五个实验室构建而成，从 1,000 笔贷款的基础模型开始，通过相关的违约、政策杠杆和经济衰退压力情景，最终形成一份财务主管可以提交给董事会的单页摘要。

✨离散事件（深度实践）：诊所患者流量模型，具有真实的每日到达时间、高峰感知人员配备和运营经理仪表板。

✨系统动力学（意识）：为什么反馈回路和延迟会导致繁荣与萧条。

✨基于代理（感知）：产品采用是如何从成千上万个个人选择中产生的。

✨毕业设计：从三个没有范式标签的开放式商业问题中选择一个，选择视角，从头开始编写规范，验证模型，并提出利益相关者可以采取行动的建议。

专为方便用户使用而设计。您无需统计学学位或编程背景。

✨ 数学教学以图画为主，公式为可选；不需要微积分。

✨ 你将学习到少数几种在商业建模中发挥主要作用的概率分布。

✨ 你不会从零开始编写 Python 代码。你会学会阅读、质疑和验证它，而这正是人工智能时代所需要的技能。

你在这里练习的工作流程也会不断累积。随着人工智能工具的改进，你今天编写的规范将生成更优质的代码，而无需修改规范本身；随着越来越多的代码来自可能存在细微错误的工具，验证习惯的价值也日益凸显。立即报名，开始构建你决策者一直以来所期待的模型吧！

你将学到什么

✨ 在不确定情况下，运用你自行构建的模型，解答实际的业务问题：资本储备、人员配置计划和产品发布。

✨ 使用规范优先的 AI 工作流程：编写结构化的模型规范，使用 Claude 或 ChatGPT 生成 Python 代码，然后迭代执行，无需从头编写代码。

✨ 通过三层验证（读取、断言、压力测试）人工智能生成的代码，以便向持怀疑态度的利益相关者证明结果的正确性。

✨ 构建包含置信区间、相关违约和经济衰退压力情景的蒙特卡罗贷款组合风险模型

✨ 构建一个离散事件诊所患者流量模型，该模型包含真实的就诊情况、高峰时段人员配备以及运营仪表盘。

✨ 识别系统动力学或基于代理的建模何时适用于特定问题，并在构建模型之前选择合适的范式。

✨ 为不确定的输入选择合适的概率分布，并避免那些悄然腐蚀商业模式的陷阱

✨ 将模拟结果以单页交付物的形式清晰呈现，使模型获得采取行动的资格。

要求

✨ 无需编程或统计学背景；你将阅读和验证代码，而不是从头开始编写，而且每个概念都采用图文并茂的方式讲解。

✨ 使用 Claude 或 ChatGPT 等 AI 聊天助手注册一个免费账户，即可根据您的需求生成代码。

✨ 浏览器；实验课程在 Udemy 内置的 Jupyter 环境中运行，因此无需安装任何软件（最后一部分介绍了本地设置，供有需要的用户参考）。

本课程适合以下人群：
⭐ 业务分析师、财务专业人士、运营主管和市场经理，他们需要在不确定性下做出决策，并且厌倦了漫长的等待工程师响应。⭐
数据分析师和数据科学家，他们希望解答历史数据无法回答的假设性问题，并将模拟技术添加到他们的分析和机器学习工具包中。⭐
希望找到一种持久可靠的人工智能工作方式的专业人士：明确问题并验证答案，而不是与人工智能竞争编写代码。⭐
本课程特意不涵盖机器学习、时间序列预测或优化。

Answer business questions under uncertainty: spec, generate, and validate simulation models with AI

What you’ll learn
⚡ Use a spec-first AI workflow: write a structured model spec, generate working Python with Claude or ChatGPT, and iterate without coding from scratch
⚡ Answer real business questions under uncertainty, from capital reserves to staffing plans to product launches, with models you build yourself
⚡ Validate AI-generated code in three layers (read, assert, stress-test) so you can defend your results to a skeptical stakeholder
⚡ Run the what-if scenarios analytics and ML can’t: simulate futures with no historical precedent and report them with confidence intervals
⚡ Build a Monte Carlo loan portfolio risk model with correlated defaults, policy levers, and recession stress scenarios
⚡ Build a discrete-event clinic patient-flow model with realistic arrivals, peak-aware staffing, and an ops manager dashboard
⚡ Choose the right paradigm for a new problem (Monte Carlo, discrete-event, system dynamics, agent-based) and the right distribution for each uncertain input
⚡ Communicate simulation results in a one-page deliverable that earns the model the right to be acted on

Requirements
❗ No programming or statistics background required. You’ll read and verify code rather than write it, and every math concept is taught picture-first, with no calculus needed.
❗ A free account with an AI chat assistant such as Claude or ChatGPT, for generating code from your specs.
❗ A web browser. Labs run in Udemy’s built-in Jupyter environment, so there’s nothing to install; the final section covers optional local setup.

Description
AI can write working code in seconds. Depending on what you do next, that’s either a threat to your career or the biggest lever you’ve ever been handed. The difference comes down to one skill: knowing exactly what to ask for, and how to verify the answer before anyone acts on it. This course teaches that skill through simulation modeling, the craft of answering business questions under uncertainty with models you can defend.

Those questions are everywhere once you look

✨ How much capital should a bank hold against next year’s loan losses?

✨ How many doctors does the clinic need on Monday morning?

✨ Should the launch budget go to ads or to influencers?

Five years ago, answering them meant hiring specialists. Generative AI changed the economics: writing simulation code used to take a senior engineer a week, and for a well-specified problem it now takes an afternoon. The bottleneck moved from typing to thinking, and thinking is where the value always lived. This course puts you on the valuable side of that line.

Already work in analytics or machine learning? Simulation fills the gap those tools leave open

✨ Analytics and ML learn from the past. They’re the right tools when yesterday’s data describes tomorrow’s world.

✨ They struggle exactly where the stakes run highest: a decision no one has made before, a shock with no precedent in the training data, a what-if question about a policy that’s never been tried.

✨ Simulation works from the other direction. You model how the system itself works (the arrivals, the queues, the defaults, the feedback loops), then run the futures that haven’t happened yet.

✨ When a stakeholder asks “what if demand doubles?” or “how bad could a severe recession get?”, a model trained on history can only extrapolate. A simulation answers, complete with confidence intervals.

That’s why simulation belongs in every analyst’s and data scientist’s toolkit, right alongside the skills you already have.

One workflow, every model. You’ll practice the same five steps on everything you build

✨Spec it. Write a structured, plain-language description of the business question, the assumptions with sources, the process, and the checks that prove the model is right.

✨Generate it. Paste your spec into Claude or ChatGPT and get working Python back.

✨Run it. Execute in a Jupyter notebook in your browser. Udemy provides the environment, so there’s nothing to install.

✨Validate it. Three layers: read the code, run the assertions, stress-test with extreme inputs.

✨Iterate. When the results raise a new question, edit the spec and regenerate.

That rhythm is the most transferable skill the course teaches, and it works the same whether the model is a risk analysis or anything else you’ll ask AI to build for you.

What you’ll build. The course covers the four major simulation paradigms, two in depth and two at awareness level

✨Monte Carlo (deep practice): a community bank’s loan portfolio risk model, built across five labs from a 1,000-loan base model through correlated defaults, policy levers, and recession stress scenarios, ending with a one-page summary a treasurer could take to the board.

✨Discrete-event (deep practice): a clinic patient-flow model with realistic time-of-day arrivals, peak-aware staffing, and an ops manager dashboard.

✨System dynamics (awareness): why feedback loops and delays produce boom and bust.

✨Agent-based (awareness): how product adoption emerges from thousands of individual choices.

✨Capstone: pick one of three open-ended business problems with no paradigm label attached, choose the lens, write the spec from scratch, validate the model, and deliver a recommendation a stakeholder could act on.

Built for accessibility. You don’t need a stats degree or a coding background

✨ Math is taught picture-first with formulas optional; nothing requires calculus.

✨ You’ll learn the handful of probability distributions that do most of the work in business modeling.

✨ You won’t write Python from scratch. You’ll learn to read it, question it, and verify it, which is exactly the skill the age of AI rewards.

The workflow you’ll practice here also compounds. The specs you write today will generate better code as the AI tools improve, without the specs needing to change, and the verification habit only grows more valuable as more of the world’s code comes from tools that can be subtly wrong. Enroll, and start building the models your decisions have been waiting for.

What you’ll learn

✨ Answer real business questions under uncertainty: capital reserves, staffing plans, and product launches, with models you build yourself

✨ Use a spec-first AI workflow: write a structured model spec, generate Python with Claude or ChatGPT, and iterate without writing code from scratch

✨ Validate AI-generated code in three layers (read, assert, stress-test) so you can defend the results to a skeptical stakeholder

✨ Build a Monte Carlo loan portfolio risk model with confidence intervals, correlated defaults, and recession stress scenarios

✨ Build a discrete-event clinic patient-flow model with realistic arrivals, peak-aware staffing, and an ops dashboard

✨ Recognize when system dynamics or agent-based modeling fits a problem, and choose the right paradigm before you build

✨ Pick the right probability distribution for an uncertain input, and avoid the traps that quietly corrupt business models

✨ Communicate simulation results in a one-page deliverable that earns the model the right to be acted on

Requirements

✨ No programming or statistics background required; you’ll read and verify code rather than write it from scratch, and every concept is taught picture-first

✨ A free account with an AI chat assistant such as Claude or ChatGPT for generating code from your specs

✨ A browser; labs run in Udemy’s built-in Jupyter environment, so there’s nothing to install (the final section covers local setup for those who want it)

Who this course is for
⭐ Business analysts, finance professionals, operations leads, and marketing managers who face decisions under uncertainty and are tired of waiting on an engineering queue.
⭐ Data analysts and data scientists who want to answer the what-if questions their historical data can’t, and add simulation to their analytics and ML toolkit.
⭐ Professionals who want a durable, defensible way to work with AI: specifying problems and verifying answers rather than competing with it at writing code.
⭐ This course deliberately does not cover machine learning, time-series forecasting, or optimization.