{"id":3213,"date":"2025-12-19T07:04:00","date_gmt":"2025-12-18T23:04:00","guid":{"rendered":"https:\/\/www.laoyulaoyu.com\/?p=3213"},"modified":"2025-12-10T17:06:09","modified_gmt":"2025-12-10T09:06:09","slug":"%e5%b8%a6%e4%bd%a0%e9%87%8f%e5%8c%96%e5%85%a5%e9%97%a8%ef%bc%9a%e5%a6%82%e4%bd%9510%e5%88%86%e9%92%9f%e6%90%ad%e5%bb%ba%e4%b8%80%e4%b8%aa%e5%bc%ba%e5%8c%96%e5%ad%a6%e4%b9%a0%e6%9c%ba%e5%99%a8%e4%ba%ba","status":"publish","type":"post","link":"https:\/\/laoyulaoyu.com\/index.php\/2025\/12\/19\/%e5%b8%a6%e4%bd%a0%e9%87%8f%e5%8c%96%e5%85%a5%e9%97%a8%ef%bc%9a%e5%a6%82%e4%bd%9510%e5%88%86%e9%92%9f%e6%90%ad%e5%bb%ba%e4%b8%80%e4%b8%aa%e5%bc%ba%e5%8c%96%e5%ad%a6%e4%b9%a0%e6%9c%ba%e5%99%a8%e4%ba%ba\/","title":{"rendered":"\u5e26\u4f60\u91cf\u5316\u5165\u95e8\uff1a\u5982\u4f5510\u5206\u949f\u642d\u5efa\u4e00\u4e2a\u5f3a\u5316\u5b66\u4e60\u673a\u5668\u4eba\uff08\u9644\u6e90\u7801\uff09"},"content":{"rendered":"\n

\u4f5c\u8005\uff1a\u8001\u4f59\u635e\u9c7c<\/a><\/p>\n\n\n\n

\u539f\u521b\u4e0d\u6613\uff0c\u8f6c\u8f7d\u8bf7\u6807\u660e\u51fa\u5904\u53ca\u539f\u4f5c\u8005\u3002<\/mark><\/strong><\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n
\u5199\u5728\u524d\u9762\u7684\u8bdd\uff1a<\/strong>\u4eca\u5929\u8fd9\u7bc7\u6587\u7ae0\u6709\u70b9\u201c\u786c\u201d\uff0c\u4f46\u6211\u4fdd\u8bc1\uff0c\u53ea\u8981\u4f60\u8010\u5fc3\u770b\u5b8c\uff0c\u4f1a\u5bf9\u201cAI\u91cf\u5316+\u91d1\u878d\u201d\u6709\u4e00\u4e2a\u5168\u65b0\u7684\u8ba4\u77e5\u3002\u54b1\u4eec\u4e0d\u6574\u865a\u7684\uff0c\u76f4\u63a5\u4e0a\u4ee3\u7801\uff0c\u6559\u4f60\u7528\u5f3a\u5316\u5b66\u4e60\u6413\u4e00\u4e2a\u4ea4\u6613\u673a\u5668\u4eba\u3002<\/pre>\n\n\n\n
\u6700\u8fd1\u603b\u6709\u7c89\u4e1d\u540e\u53f0\u95ee\u6211\uff1a\u201c\u8001\u4f59\uff0c\u6211\u770b\u90a3\u4e2a\u91cf\u5316\u4ea4\u6613\u597d\u50cf\u5f88\u8d5a\u94b1\uff0c\u4f46\u6211\u4e0d\u4f1a\u5199\u590d\u6742\u7684\u6570\u5b66\u516c\u5f0f\uff0c\u80fd\u4e0d\u80fd\u8ba9AI\u81ea\u5df1\u53bb\u5b66\u600e\u4e48\u7092\u80a1\uff1f\u201d<\/p>\n\n\n\n
\u95ee\u5f97\u597d\uff01\u8fd9\u5c31\u597d\u6bd4\u4f60\u95ee\uff1a\u201c\u6211\u4e0d\u60f3\u5b66\u5f00\u8f66\uff0c\u80fd\u4e0d\u80fd\u641e\u8f86\u81ea\u52a8\u9a7e\u9a76\u7684\u8f66\uff1f\u201d<\/p>\n\n\n\n
\u7b54\u6848\u662f\uff1a\u53ef\u4ee5\uff0c\u4f46\u4f60\u5f97\u5148\u5b66\u4f1a\u600e\u4e48\u9020\u8f66\u3002<\/strong><\/p>\n\n\n\n
\u4eca\u5929\uff0c\u54b1\u4eec\u5c31\u6765\u804a\u804a\u91cf\u5316\u9886\u57df\u6700\u6027\u611f\u3001\u4e5f\u6700\u786c\u6838\u7684\u6280\u672f\u2014\u2014\u5f3a\u5316\u5b66\u4e60\uff08Reinforcement Learning, RL\uff09<\/strong>\u3002\u522b\u88ab\u8fd9\u4e2a\u540d\u8bcd\u5413\u5230\u4e86\uff0c\u5176\u5b9e\u5b83\u7684\u539f\u7406\u8ddf\u9a6f\u72d7\u5dee\u4e0d\u591a\u3002\u6211\u4eec\u5c06\u4e00\u6b65\u6b65\u642d\u5efa\u4e00\u4e2aAI\uff0c\u8ba9\u5b83\u5728\u80a1\u5e02\u7684\u8d77\u8d77\u4f0f\u4f0f\u4e2d\uff0c\u81ea\u5df1\u5b66\u4f1a\u4ec0\u4e48\u65f6\u5019\u4e70\uff0c\u4ec0\u4e48\u65f6\u5019\u5356\u3002<\/p>\n\n\n\n
\u4e00\u3001 \u4ec0\u4e48\u662f\u5f3a\u5316\u5b66\u4e60\uff1f<\/strong><\/h2>\n\n\n\n
\u4f60\u53ef\u4ee5\u628a\u8bad\u7ec3AI\u4ea4\u6613\u5458\u60f3\u8c61\u6210\u8bad\u7ec3\u4e00\u53ea\u521a\u8fdb\u5bb6\u95e8\u7684\u5c0f\u72d7\uff1a<\/p>\n\n\n\n
    \n
  • \u73af\u5883 (Environment)<\/strong>\uff1a\u8fd9\u5c31\u662f\u80a1\u5e02\u3002\u5145\u6ee1\u4e86\u8bf1\u60d1\uff08\u66b4\u6da8\uff09\u548c\u9677\u9631\uff08\u66b4\u8dcc\uff09\u3002<\/li>\n\n\n\n
  • \u667a\u80fd\u4f53 (Agent)<\/strong>\uff1a\u8fd9\u5c31\u662f\u4f60\u7684AI\u673a\u5668\u4eba\uff0c\u6216\u8005\u90a3\u53ea\u5c0f\u72d7\u3002<\/li>\n\n\n\n
  • \u52a8\u4f5c (Action)<\/strong>\uff1a\u5c0f\u72d7\u80fd\u505a\u7684\u52a8\u4f5c\u2014\u2014\u6bd4\u5982\u201c\u5750\u4e0b\u201d\u3001\u201c\u63e1\u624b\u201d\u3002\u5728\u80a1\u5e02\u91cc\uff0c\u5c31\u662f\u201c\u4e70\u5165\u201d\u3001\u201c\u5356\u51fa\u201d\u6216\u8005\u201c\u8eba\u5e73\uff08\u6301\u6709\uff09\u201d\u3002<\/li>\n\n\n\n
  • \u5956\u52b1 (Reward)<\/strong>\uff1a\u8fd9\u662f\u6838\u5fc3\uff01\u5c0f\u72d7\u505a\u5bf9\u4e86\u52a8\u4f5c\uff0c\u4f60\u7ed9\u5b83\u8089\u5e72\uff1b\u505a\u9519\u4e86\uff0c\u4f60\u751a\u81f3\u53ef\u80fd\u8981\u9a82\u5b83\u4e24\u53e5\u3002\u5728\u4ea4\u6613\u91cc\uff0c\u8d5a\u4e86\u94b1\u5c31\u662f\u8089\u5e72\uff08\u6b63\u5956\u52b1\uff09\uff0c\u4e8f\u4e86\u94b1\u5c31\u662f\u60e9\u7f5a\uff08\u8d1f\u5956\u52b1\uff09\u3002<\/li>\n<\/ul>\n\n\n\n
    \"\"
    \u6211\u4eec\u76ee\u6807\u662f\u6784\u5efa\u4e00\u4e2a\u80fd\u81ea\u6211\u8fdb\u5316\u7684AI\u4ea4\u6613\u5458<\/figcaption><\/figure>\n\n\n\n
    \u901a\u8fc7\u6210\u5343\u4e0a\u4e07\u6b21\u7684\u5c1d\u8bd5\uff0cAI\u4f1a\u6162\u6162\u603b\u7ed3\u51fa\u4e00\u5957\u89c4\u5f8b\uff1a\u201c\u54e6\uff0c\u539f\u6765\u5728\u8fd9\u4e2a\u6307\u6807\u91d1\u53c9\u7684\u65f6\u5019\u4e70\u5165\uff0c\u5927\u6982\u7387\u80fd\u5403\u5230\u8089\u5e72\uff01\u201d<\/p>\n\n\n\n
    \u8fd9\u5c31\u53eb\u7b56\u7565\u4f18\u5316<\/strong>\u3002<\/p>\n\n\n\n
    \u4e3a\u4e86\u8ba9\u5927\u5bb6\u66f4\u76f4\u89c2\u5730\u7406\u89e3\uff0c\u6211\u505a\u4e86\u4e00\u4e2a\u5bf9\u6bd4\u8868\uff1a<\/p>\n\n\n\n
    \u7279\u6027<\/th>\u4f20\u7edf\u91cf\u5316\u7b56\u7565<\/th>\u5f3a\u5316\u5b66\u4e60 (RL) \u7b56\u7565<\/th><\/tr>
    \u51b3\u7b56\u903b\u8f91<\/strong><\/td>\u6b7b\u8bb0\u786c\u80cc\uff08\u5982\u679cA>B\uff0c\u5219\u4e70\u5165\uff09<\/td>\u968f\u673a\u5e94\u53d8\uff08\u6839\u636e\u76d8\u9762\u611f\u89c9\u52a8\u6001\u8c03\u6574\uff09<\/td><\/tr>
    \u9002\u5e94\u6027<\/strong><\/td>\u5e02\u573a\u53d8\u4e86\uff0c\u7b56\u7565\u5c31\u5931\u6548<\/td>\u53ef\u4ee5\u6301\u7eed\u5b66\u4e60\uff0c\u81ea\u6211\u8fdb\u5316<\/td><\/tr>
    \u5f00\u53d1\u96be\u5ea6<\/strong><\/td>\u9700\u8981\u7cbe\u901a\u91d1\u878d\u6307\u6807<\/td>\u9700\u8981\u7cbe\u901a\u201c\u8bbe\u8ba1\u5956\u52b1\u673a\u5236\u201d<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
    \u4e8c\u3001 \u51c6\u5907\u5de5\u4f5c\uff1a\u642d\u597d\u4f60\u7684\u5175\u5668\u5e93<\/strong><\/h2>\n\n\n\n
    \u5de5\u6b32\u5584\u5176\u4e8b\uff0c\u5fc5\u5148\u5229\u5176\u5668\u3002\u54b1\u4eec\u8fd9\u6b21\u4e3b\u8981\u7528\u5230 Python\uff0c\u4ee5\u53ca\u51e0\u4e2a\u6838\u5fc3\u7684\u795e\u5668\u3002\u8001\u4f59\u5efa\u8bae\u4f60\u7528 Python 3.8 \u4ee5\u4e0a\u7684\u7248\u672c\u3002<\/p>\n\n\n\n
    \u6253\u5f00\u4f60\u7684\u7ec8\u7aef\uff08Terminal\uff09\uff0c\u628a\u4e0b\u9762\u8fd9\u4e9b\u5e93\u88c5\u4e0a\uff1a<\/p>\n\n\n\n
    pip install numpy pandas gymnasium stable-baselines3 yfinance matplotlib<\/code><\/pre>\n\n\n\n
    \u8fd9\u91cc\u91cd\u70b9\u4ecb\u7ecd\u4e24\u4f4d\u4e3b\u89d2\uff1a<\/p>\n\n\n\n
      \n
    • Gymnasium<\/strong>\uff1a\u8fd9\u662fOpenAI Gym\u7684\u5347\u7ea7\u7248\uff0c\u7528\u6765\u642d\u5efa\u6211\u4eec\u7684\u201c\u6a21\u62df\u4ea4\u6613\u6240\u201d\u3002<\/li>\n\n\n\n
    • Stable-Baselines3<\/strong>\uff1a\u8fd9\u662f\u91cf\u5316\u5708\u7684\u201c\u745e\u58eb\u519b\u5200\u201d\uff0c\u91cc\u9762\u5c01\u88c5\u597d\u4e86\u6700\u9876\u7ea7\u7684\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\uff08\u6bd4\u5982PPO\uff09\uff0c\u7701\u5f97\u4f60\u81ea\u5df1\u53bb\u5199\u90a3\u51e0\u5343\u884c\u5e95\u5c42\u4ee3\u7801\u3002<\/li>\n<\/ul>\n\n\n\n
      \u4e09\u3001 \u5b9e\u6218\u7b2c\u4e00\u6b65\uff1a\u641e\u6570\u636e<\/strong><\/h2>\n\n\n\n
      \u5de7\u5987\u96be\u4e3a\u65e0\u7c73\u4e4b\u708a\u3002\u54b1\u4eec\u5148\u7528 yfinance<\/code> \u628a\u82f9\u679c\u516c\u53f8 (AAPL) \u7684\u80a1\u7968\u6570\u636e\u6252\u62c9\u4e0b\u6765\u3002<\/p>\n\n\n\n
      import yfinance as yf\nimport pandas as pd\n\n# \u4e0b\u8f7d\u82f9\u679c\u516c\u53f8\u7684\u6570\u636e\uff0c\u4ece2020\u5e74\u52302024\u5e74\ndef get_data(ticker='AAPL'):\n    data = yf.download(ticker, start='2020-01-01', end='2024-01-01')\n    # \u7b80\u5355\u7684\u9884\u5904\u7406\uff0c\u6211\u4eec\u53ea\u9700\u8981\u6536\u76d8\u4ef7\n    data = data[['Close']]\n    return data\n\ndf = get_data()\nprint(df.head())<\/code><\/pre>\n\n\n\n
      \u8001\u4f59\u63d0\u793a\uff1a\u771f\u5b9e\u505a\u91cf\u5316\u65f6\uff0c\u6570\u636e\u6e05\u6d17\u80fd\u5360\u4f6080%\u7684\u65f6\u95f4\u3002\u8fd9\u91cc\u4e3a\u4e86\u6f14\u793a\u65b9\u4fbf\uff0c\u6211\u4eec\u5047\u8bbe\u6570\u636e\u662f\u5b8c\u7f8e\u7684\u3002<\/em><\/p>\n\n\n\n
      \u56db\u3001 \u6838\u5fc3\u73af\u8282\uff1a\u642d\u5efa\u201c\u7ec3\u529f\u623f\u201d<\/strong><\/h2>\n\n\n\n
      \u8fd9\u662f\u6700\u5173\u952e\u7684\u4e00\u6b65\uff01\u6211\u4eec\u9700\u8981\u544a\u8bc9AI\uff1a\u4f60\u770b\u5f97\u5230\u4ec0\u4e48\uff08\u72b6\u6001\uff09\uff0c\u4f60\u80fd\u505a\u4ec0\u4e48\uff08\u52a8\u4f5c\uff09\uff0c\u4ee5\u53ca\u600e\u4e48\u7b97\u8d62\uff08\u5956\u52b1\uff09\u3002<\/p>\n\n\n\n
      \u6211\u4eec\u9700\u8981\u7ee7\u627f gym.Env<\/code> \u6765\u5199\u4e00\u4e2a\u81ea\u5b9a\u4e49\u73af\u5883\u3002\u4ee3\u7801\u6709\u70b9\u957f\uff0c\u4f46\u903b\u8f91\u5f88\u6e05\u6670\uff0c\u8ddf\u7740\u8001\u4f59\u7684\u6ce8\u91ca\u770b\uff1a<\/p>\n\n\n\n
      import gymnasium as gym\nimport numpy as np\nfrom gymnasium import spaces\n\nclass StockTradingEnv(gym.Env):\n    \"\"\"\n    \u81ea\u5b9a\u4e49\u80a1\u7968\u4ea4\u6613\u73af\u5883\n    \"\"\"\n    def __init__(self, df, initial_balance=10000):\n        super(StockTradingEnv, self).__init__()\n        \n        self.df = df\n        self.initial_balance = initial_balance\n        \n        # \u3010\u52a8\u4f5c\u7a7a\u95f4\u3011\uff1a3\u4e2a\u52a8\u4f5c\n        # 0: \u5356\u51fa, 1: \u6301\u6709, 2: \u4e70\u5165\n        self.action_space = spaces.Discrete(3)\n        \n        # \u3010\u89c2\u5bdf\u7a7a\u95f4\u3011\uff1aAI\u80fd\u770b\u5230\u7684\u6570\u636e\n        # \u8fd9\u91cc\u4e3a\u4e86\u7b80\u5355\uff0c\u6211\u4eec\u5047\u8bbeAI\u80fd\u770b\u5230\uff1a\n        # [\u5f53\u524d\u80a1\u4ef7, \u8d26\u6237\u4f59\u989d, \u6301\u6709\u7684\u80a1\u6570, \u76c8\u4e8f\u6bd4\u4f8b]\n        self.observation_space = spaces.Box(\n            low=0, high=np.inf, shape=(6,), dtype=np.float32\n        )\n        \n    def reset(self, seed=None, options=None):\n        # \u91cd\u7f6e\u73af\u5883\uff0c\u6bd4\u5982\u5f00\u59cb\u65b0\u4e00\u8f6e\u8bad\u7ec3\n        super().reset(seed=seed)\n        \n        self.balance = self.initial_balance # \u521d\u59cb\u8d44\u91d1\n        self.shares_held = 0  # \u6301\u4ed3\u80a1\u6570\n        self.current_step = 0 # \u5f53\u524d\u5929\u6570\n        self.total_asset = self.initial_balance\n        \n        return self._next_observation(), {}\n\n    def _next_observation(self):\n        # \u83b7\u53d6\u5f53\u524d\u7684\u72b6\u6001\u6570\u636e\u7ed9AI\u770b\n        # \u771f\u5b9e\u573a\u666f\u91cc\uff0c\u8fd9\u91cc\u901a\u5e38\u4f1a\u52a0\u5165RSI, MACD\u7b49\u6280\u672f\u6307\u6807\n        obs = np.array([\n            self.df.iloc[self.current_step]['Close'].item(),\n            self.balance,\n            self.shares_held,\n            0, 0, 0 # \u9884\u7559\u7ed9\u5176\u4ed6\u6307\u6807\u7684\u4f4d\u7f6e\n        ])\n        return obs.astype(np.float32)\n\n    def step(self, action):\n        # AI\u6267\u884c\u4e00\u4e2a\u52a8\u4f5c\uff0c\u73af\u5883\u7ed9\u53cd\u9988\n        \n        current_price = self.df.iloc[self.current_step]['Close'].item()\n        \n        # \u6267\u884c\u4e70\u5356\u903b\u8f91\n        if action == 2: # \u4e70\u5165\n            # \u7b80\u5355\u7b56\u7565\uff1a\u6bcf\u6b21\u6ee1\u4ed3\u68ad\u54c8\uff08\u4ec5\u4f9b\u6559\u5b66\uff01\uff09\n            # \u5b9e\u9645\u4e2d\u8981\u8003\u8651\u4ed3\u4f4d\u7ba1\u7406\n            total_possible = self.balance \/\/ current_price\n            if total_possible > 0:\n                self.shares_held += total_possible\n                self.balance -= total_possible * current_price\n                \n        elif action == 0: # \u5356\u51fa\n            # \u6e05\u4ed3\u5356\u51fa\n            if self.shares_held > 0:\n                self.balance += self.shares_held * current_price\n                self.shares_held = 0\n        \n        # \u66f4\u65b0\u603b\u8d44\u4ea7 = \u73b0\u91d1 + \u80a1\u7968\u5e02\u503c\n        new_total_asset = self.balance + (self.shares_held * current_price)\n        \n        # \u3010\u5956\u52b1\u51fd\u6570\u3011\uff1a\u6700\u6838\u5fc3\u7684\u90e8\u5206\uff01\n        # \u4eca\u5929\u7684\u5956\u52b1 = \u4eca\u5929\u7684\u603b\u8d44\u4ea7 - \u6628\u5929\u7684\u603b\u8d44\u4ea7\n        reward = new_total_asset - self.total_asset\n        self.total_asset = new_total_asset\n        \n        self.current_step += 1\n        \n        # \u5224\u65ad\u662f\u5426\u7ed3\u675f\uff08\u94b1\u82b1\u5149\u4e86\uff0c\u6216\u8005\u6570\u636e\u8dd1\u5b8c\u4e86\uff09\n        done = self.current_step >= len(self.df) - 1\n        \n        return self._next_observation(), reward, done, False, {}\n\n    def render(self):\n        # \u6253\u5370\u5f53\u524d\u72b6\u6001\uff0c\u7528\u4e8e\u8c03\u8bd5\n        print(f'Step: {self.current_step}, Asset: {self.total_asset}')\n<\/code><\/pre>\n\n\n\n
      \u8fd9\u4e2a StockTradingEnv<\/code> \u5c31\u662f\u6211\u4eec\u7684\u7ec3\u529f\u623f\u3002\u6ce8\u610f\u90a3\u4e2a Reward<\/strong> \u7684\u8ba1\u7b97\uff0c\u8fd9\u91cc\u662f\u6700\u7b80\u5355\u7684\u201c\u8d5a\u4e86\u591a\u5c11\u5c31\u662f\u591a\u5c11\u5206\u201d\u3002<\/p>\n\n\n\n
      \u8001\u4f59\u6572\u9ed1\u677f\uff1a<\/strong> \u5728\u5b9e\u9645\u751f\u4ea7\u4e2d\uff0c\u5355\u7eaf\u7528\u5229\u6da6\u505a\u5956\u52b1\u5f88\u5371\u9669\uff0cAI\u4e3a\u4e86\u8ffd\u6c42\u9ad8\u6536\u76ca\u4f1a\u901a\u8fc7\u9891\u7e41\u4ea4\u6613\u6765\u201c\u8d4c\u535a\u201d\u3002\u901a\u5e38\u6211\u4eec\u8fd8\u4f1a\u51cf\u53bb\u4ea4\u6613\u624b\u7eed\u8d39\uff0c\u751a\u81f3\u5f15\u5165\u590f\u666e\u6bd4\u7387\uff08Sharpe Ratio\uff09\u6765\u5956\u52b1\u98ce\u9669\u63a7\u5236\u5f97\u597d\u7684\u64cd\u4f5c\u3002<\/p>\n\n\n\n
      \u4e94\u3001 \u53ec\u5524AI\u5927\u795e\uff1aPPO\u7b97\u6cd5<\/strong><\/h2>\n\n\n\n
      \u73af\u5883\u642d\u597d\u4e86\uff0c\u63a5\u4e0b\u6765\u6211\u4eec\u4e0d\u9700\u8981\u81ea\u5df1\u5199\u795e\u7ecf\u7f51\u7edc\uff0c\u76f4\u63a5\u8c03\u7528 Stable-Baselines3<\/code> \u91cc\u7684 PPO \u7b97\u6cd5\u3002<\/p>\n\n\n\n
      \u4e3a\u4ec0\u4e48\u9009 PPO\uff1f\u56e0\u4e3a\u5b83\u7a33\u3002\u5728\u5f3a\u5316\u5b66\u4e60\u754c\uff0c\u5b83\u5c31\u50cf\u662f\u4e30\u7530\u5361\u7f57\u62c9\uff0c\u867d\u7136\u4e0d\u662f\u6700\u5feb\u7684\uff0c\u4f46\u662f\u6700\u76ae\u5b9e\u8010\u7528\uff0c\u4e0d\u5bb9\u6613\u7ffb\u8f66\u3002<\/p>\n\n\n\n
      from stable_baselines3 import PPO\n\n# 1. \u5b9e\u4f8b\u5316\u73af\u5883\nenv = StockTradingEnv(df)\n\n# 2. \u5b9a\u4e49\u6a21\u578b\n# MlpPolicy \u662f\u6307\u4f7f\u7528\u591a\u5c42\u611f\u77e5\u673a\uff08\u7b80\u5355\u7684\u795e\u7ecf\u7f51\u7edc\uff09\nmodel = PPO(\"MlpPolicy\", env, verbose=1, learning_rate=0.0003)\n\n# 3. \u5f00\u59cb\u4fee\u70bc\uff01\nprint(\"\u6b63\u5728\u8bad\u7ec3\u673a\u5668\u4eba\uff0c\u8bf7\u7a0d\u5019...\")\nmodel.learn(total_timesteps=100000) # \u8bad\u7ec310\u4e07\u6b65\nprint(\"\u8bad\u7ec3\u5b8c\u6210\uff01\")<\/code><\/pre>\n\n\n\n
      \u5f53\u4f60\u8fd0\u884c\u8fd9\u884c\u4ee3\u7801\u65f6\uff0c\u4f60\u7684CPU\/GPU\u5c31\u5f00\u59cb\u75af\u72c2\u8fd0\u8f6c\u4e86\u3002\u8fd9\u5c31\u50cf\u628a\u4e00\u4e2a\u65b0\u80a1\u6c11\u5173\u5728\u5c0f\u9ed1\u5c4b\u91cc\uff0c\u8ba9\u4ed6\u5bf9\u7740\u8fc7\u53bb4\u5e74\u7684K\u7ebf\u56fe\u53cd\u590d\u6a21\u62df\u4ea4\u661310\u4e07\u6b21\u3002<\/p>\n\n\n\n
      \u516d\u3001 \u662f\u9aa1\u5b50\u662f\u9a6c\uff0c\u62c9\u51fa\u6765\u905b\u905b<\/strong><\/h2>\n\n\n\n
      \u8bad\u7ec3\u5b8c\u4e86\uff0c\u6548\u679c\u548b\u6837\uff1f\u5343\u4e07\u522b\u62ff\u8bad\u7ec3\u8fc7\u7684\u6570\u636e\u53bb\u6d4b\u8bd5\uff0c\u90a3\u662f\u4f5c\u5f0a\uff08Overfitting\uff0c\u8fc7\u62df\u5408\uff09\u3002<\/p>\n\n\n\n
      \u4f60\u9700\u8981\u4e00\u6bb5AI\u4ece\u672a\u89c1\u8fc7\u7684\u6570\u636e\u6765\u505a\u56de\u6d4b\u3002<\/p>\n\n\n\n
      # \u7b80\u5355\u7684\u9a8c\u8bc1\u903b\u8f91\nobs, _ = env.reset()\nfor i in range(100):\n    action, _states = model.predict(obs)\n    obs, rewards, done, truncated, info = env.step(action)\n    env.render()\n    if done:\n        break<\/code><\/pre>\n\n\n\n
      \u5982\u679c\u4f60\u7684\u8f93\u51fa\u7ed3\u679c\u91cc\uff0c\u8d44\u4ea7\u4e00\u76f4\u5728\u7a33\u6b65\u589e\u957f\uff0c\u606d\u559c\u4f60\uff0c\u4f60\u5165\u95e8\u4e86\uff01\u4f46\u5982\u679c\u8d44\u4ea7\u5f52\u96f6\u4e86\uff0c\u4e5f\u522b\u7070\u5fc3\uff0c\u8fd9\u624d\u662f\u5e38\u6001\u3002<\/p>\n\n\n\n
      \u4e03\u3001 \u907f\u5751\u6307\u5357<\/strong><\/h2>\n\n\n\n
      \u4ee3\u7801\u8dd1\u901a\u4e86\uff0c\u662f\u4e0d\u662f\u89c9\u5f97\u81ea\u5df1\u660e\u5929\u5c31\u80fd\u62f3\u6253\u5df4\u83f2\u7279\uff0c\u811a\u8e22\u7d22\u7f57\u65af\u4e86\uff1f<\/p>\n\n\n\n
      \u9192\u9192\uff01<\/strong> \u8fd9\u91cc\u6709\u51e0\u4e2a\u5de8\u5751\uff0c\u8db3\u4ee5\u8ba9\u4f60\u4e8f\u5f97\u88e4\u8869\u90fd\u4e0d\u5269\uff1a<\/p>\n\n\n\n
        \n
      1. \u624b\u7eed\u8d39\u731b\u5982\u864e<\/strong>\uff1a\u4e0a\u9762\u7684\u4ee3\u7801\u6ca1\u7b97\u624b\u7eed\u8d39\u3002\u5982\u679c\u4f60\u4e0d\u628a\u6bcf\u7b14\u4ea4\u6613\u5343\u5206\u4e4b\u4e00\u751a\u81f3\u5343\u5206\u4e4b\u4e09\u7684\u6210\u672c\u7b97\u8fdb Reward \u91cc\uff0cAI \u5c31\u4f1a\u53d8\u6210\u9ad8\u9891\u4ea4\u6613\u72c2\u9b54\uff0c\u6700\u540e\u94b1\u5168\u4ea4\u7ed9\u5238\u5546\u4e86\u3002<\/li>\n\n\n\n
      2. \u672a\u6765\u51fd\u6570 (Look-ahead Bias)<\/strong>\uff1a\u5343\u4e07\u522b\u5728\u4eca\u5929\u7684\u6570\u636e\u91cc\u585e\u8fdb\u660e\u5929\u7684\u6536\u76d8\u4ef7\u3002\u8fd9\u662f\u65b0\u624b\u6700\u5bb9\u6613\u72af\u7684\u9519\uff0c\u56de\u6d4b\u6536\u76ca\u7387200%\uff0c\u5b9e\u76d8\u76f4\u63a5\u8170\u65a9\u3002<\/li>\n\n\n\n
      3. \u5e78\u5b58\u8005\u504f\u5dee<\/strong>\uff1a\u5982\u679c\u4f60\u53ea\u7528\u73b0\u5728\u8fd8\u5728\u4e0a\u5e02\u7684\u516c\u53f8\u6570\u636e\u8bad\u7ec3\uff0cAI \u5c31\u6c38\u8fdc\u5b66\u4e0d\u4f1a\u201c\u516c\u53f8\u4f1a\u9000\u5e02\u201d\u8fd9\u4ef6\u4e8b\u3002<\/li>\n<\/ol>\n\n\n\n
        \u516b\u3001 \u89c2\u70b9\u603b\u7ed3<\/strong><\/h2>\n\n\n\n
        \u4eca\u5929\u6211\u4eec\u7528\u4e0d\u5230100\u884c\u4ee3\u7801\uff0c\u6784\u5efa\u4e86\u4e00\u4e2a\u6700\u57fa\u7840\u7684RL\u4ea4\u6613\u7cfb\u7edf\u3002\u4f46\u8fd9\u53ea\u662f\u4e07\u91cc\u957f\u5f81\u7b2c\u4e00\u6b65\u3002<\/p>\n\n\n\n
        \u5982\u679c\u4f60\u60f3\u8fdb\u9636\uff0c\u53ef\u4ee5\u5c1d\u8bd5\uff1a<\/p>\n\n\n\n