复杂网络环境下AI Agent的局部交互与全局涌现行为分析

import networkx as nx
import numpy as np
import random
import matplotlib.pyplot as plt

# 构建小世界网络
G = nx.watts_strogatz_graph(n=20, k=4, p=0.3)

# 定义Agent类
class Agent:
    def __init__(self, actions=[0, 1]):  # 0=合作，1=背叛
        self.q_table = {a: 0.0 for a in actions}
        self.actions = actions
        self.epsilon = 0.2  # 探索率
        self.alpha = 0.1    # 学习率
        self.gamma = 0.9    # 折扣因子

    def choose_action(self):
        if random.random() < self.epsilon:
            return random.choice(self.actions)
        return max(self.q_table, key=self.q_table.get)

    def update(self, action, reward):
        predict = self.q_table[action]
        target = reward + self.gamma * max(self.q_table.values())
        self.q_table[action] += self.alpha * (target - predict)

# 初始化Agent
agents = {node: Agent() for node in G.nodes}

# 模拟交互
def interaction(a1, a2):
    # 定义奖励规则（囚徒困境）
    if a1 == 0 and a2 == 0:
        return (3, 3)  # 双方合作
    elif a1 == 0 and a2 == 1:
        return (0, 5)  # A合作 B背叛
    elif a1 == 1 and a2 == 0:
        return (5, 0)  # A背叛 B合作
    else:
        return (1, 1)  # 双方背叛

# 训练过程
episodes = 200
coop_rate = []

for ep in range(episodes):
    coop_count = 0
    for node in G.nodes:
        neighbors = list(G.neighbors(node))
        if not neighbors: continue
        opponent = random.choice(neighbors)

        a1 = agents[node].choose_action()
        a2 = agents[opponent].choose_action()

        r1, r2 = interaction(a1, a2)

        agents[node].update(a1, r1)
        agents[opponent].update(a2, r2)

        if a1 == 0: coop_count += 1

    coop_rate.append(coop_count / len(G.nodes))

# 可视化合作率随时间变化
plt.plot(coop_rate)
plt.xlabel("Episodes")
plt.ylabel("Cooperation Rate")
plt.title("Self-Organized Learning in Complex Network")
plt.show()