融合强化学习与知识图谱的AI Agent决策支持框架

数据源 → 感知层 → 认知层 → 决策层 → 执行层 → 环境反馈

import numpy as np
import random

# 环境定义：库存管理
class InventoryEnv:
    def __init__(self, max_inventory=50, demand_mean=20):
        self.max_inventory = max_inventory
        self.demand_mean = demand_mean
        self.reset()

    def reset(self):
        self.inventory = self.max_inventory // 2
        return self.inventory

    def step(self, action):
        # action = 当日进货量
        self.inventory += action
        demand = np.random.poisson(self.demand_mean)  # 随机需求
        sales = min(self.inventory, demand)
        self.inventory -= sales

        # 奖励函数 = 销售利润 - 库存成本
        reward = sales * 10 - self.inventory * 2
        return self.inventory, reward

# 简单Q-learning Agent
class QLearningAgent:
    def __init__(self, state_size, action_size, lr=0.1, gamma=0.9, epsilon=0.2):
        self.q_table = np.zeros((state_size, action_size))
        self.lr = lr
        self.gamma = gamma
        self.epsilon = epsilon
        self.action_size = action_size

    def choose_action(self, state):
        if random.uniform(0,1) < self.epsilon:
            return random.randint(0, self.action_size-1)
        return np.argmax(self.q_table[state])

    def learn(self, state, action, reward, next_state):
        predict = self.q_table[state, action]
        target = reward + self.gamma * np.max(self.q_table[next_state])
        self.q_table[state, action] += self.lr * (target - predict)

# 训练过程
env = InventoryEnv()
agent = QLearningAgent(state_size=51, action_size=6)  # 动作为进货0~5单位

episodes = 200
for ep in range(episodes):
    state = env.reset()
    total_reward = 0
    for _ in range(30):  # 模拟30天
        action = agent.choose_action(state)
        next_state, reward = env.step(action)
        agent.learn(state, action, reward, next_state)
        state = next_state
        total_reward += reward
    if ep % 20 == 0:
        print(f"Episode {ep}, Total Reward: {total_reward}")

print("训练完成，Q表前5行：")
print(agent.q_table[:5])