AI Agent在复杂环境中的无模型决策机制设计与实现

import numpy as np
import gym

# 创建环境
env = gym.make("FrozenLake-v1", is_slippery=False)  # 冰湖环境，非随机滑动
n_states = env.observation_space.n
n_actions = env.action_space.n

# 初始化Q表
Q_table = np.zeros((n_states, n_actions))

# 超参数
alpha = 0.8      # 学习率
gamma = 0.95     # 折扣因子
epsilon = 1.0    # 探索率
epsilon_min = 0.01
epsilon_decay = 0.995
episodes = 1000

# Q-Learning算法
for ep in range(episodes):
    state = env.reset()[0]
    done = False

    while not done:
        # ε-贪婪选择动作
        if np.random.rand() < epsilon:
            action = env.action_space.sample()
        else:
            action = np.argmax(Q_table[state, :])

        # 执行动作
        next_state, reward, done, _, _ = env.step(action)

        # Q值更新
        Q_table[state, action] += alpha * (
            reward + gamma * np.max(Q_table[next_state, :]) - Q_table[state, action]
        )

        state = next_state

    # 动态调整epsilon
    if epsilon > epsilon_min:
        epsilon *= epsilon_decay

# 测试训练好的Agent
state = env.reset()[0]
env.render()
done = False
total_reward = 0

while not done:
    action = np.argmax(Q_table[state, :])
    next_state, reward, done, _, _ = env.step(action)
    env.render()
    state = next_state
    total_reward += reward

print("智能体总奖励:", total_reward)