一、框架概览
Mesa 是 Python 语言实现的开源 ABM (基于个体模型) 框架，自 2014 年首次发布以来，已成为复杂系统动态模拟领域重要工具。Mesa 3.2.0 是最新系列，对原有功能进行了重构与提升。
Mesa 三大核心模块
ABM 核心组件：Model 、Agent 、Space 、Time
数据收集与实验设计：DataCollector 、参数扫描
可视化系统：SolaraViz 编写的 Web UI
二、ABM 核心组件详解
1. Model 类
作为模型的主类，包括以下内容:
agent 初始化
space 初始化
random seed 和 scheduler
from mesa import Modelfrom mesa.space import MultiGridfrom my_agent import MyAgent
class MyModel(Model): def __init__(self, N, width, height): super().__init__() self.grid = MultiGrid(width, height, True) MyAgent.create_agents(self, N)
for agent in self.agents: x = self.random.randrange(self.grid.width) y = self.random.randrange(self.grid.height) self.grid.place_agent(agent, (x, y))
def step(self): self.agents.shuffle_do("step")
2. Agent 类
from mesa import Agent
class MyAgent(Agent): def __init__(self, model): super().__init__(model) self.energy = 10
def step(self): self.energy -= 1 if self.energy <= 0: self.remove() else: self.move()
def move(self): possible_steps = self.model.grid.get_neighborhood( self.pos, moore=True, include_center=False ) new_position = self.random.choice(possible_steps) self.model.grid.move_agent(self, new_position)
3. AgentSet
wealthy = model.agents.select(lambda a: a.wealth > 1000)avg_wealth = wealthy.agg("wealth", func=np.mean)
4. Spaces
from mesa.space import ContinuousSpacespace = ContinuousSpace(100, 100, torus=True)agent = MyAgent(0, model)space.place_agent(agent, (5.0, 5.0))
5. 时间控制
for _ in range(100): model.step()
三、数据收集与实验
1. DataCollector
from mesa.datacollection import DataCollector
self.datacollector = DataCollector( model_reporters={"Step": lambda m: m.steps}, agent_reporters={"Energy": "energy"})
2. 参数扫描
from mesa.batchrunner import batch_run
results = batch_run( MyModel, parameters={"N": range(10, 51, 10)}, fixed_parameters={"width": 10, "height": 10}, iterations=5, max_steps=20)
四、可视化
SolaraViz 示例
from mesa.visualization import SolaraViz, make_space_component, make_plot_component
viz = SolaraViz( model=MyModel, components=[ make_space_component(portrayal_function), make_plot_component(["Step", "Energy"]), ], model_params={"N": 10, "width": 10, "height": 10})
五、完整示例
积分薪资模型
from mesa import Agent, Modelfrom mesa.datacollection import DataCollectorimport matplotlib.pyplot as pltimport pandas as pd
class WorkerAgent(Agent): def __init__(self, model): super().__init__(model) self.salary = 100 self.savings = 0
def step(self): self.savings += self.salary * 0.2
class WorkerModel(Model): def __init__(self, N):
super().__init__() self.datacollector = DataCollector( model_reporters={"TotalSavings": self.compute_total_savings}, agent_reporters={"Savings": "savings"} ) WorkerAgent.create_agents(self, N)
def compute_total_savings(self): return sum(agent.savings for agent in self.agents)
def step(self): self.datacollector.collect(self) self.agents.do("step")
model = WorkerModel(10)for _ in range(20): model.step()
results = model.datacollector.get_model_vars_dataframe()results.plot()plt.title("Total Savings Over Time")plt.xlabel("Step")plt.ylabel("Total Savings")plt.grid(True)plt.show()
六、社区与扩展推荐
| |
|---|
mesa-geo | |
mesa-frames | |
mesa-examples | |
mesa-tutorials | |
注：本文由AI助手改写自Hoeven, E. ter, Kwakkel, J., Hess, V., Pike, T., Wang, B., Rht, & Kazil, J. (2025). Mesa 3: Agent-based modeling with Python in 2025. Journal of Open Source Software, 10(107), 7668.
