{"id":1877,"date":"2025-02-04T07:55:00","date_gmt":"2025-02-03T23:55:00","guid":{"rendered":"https:\/\/blog.laoyulaoyu.top\/?p=1877"},"modified":"2025-01-08T11:59:19","modified_gmt":"2025-01-08T03:59:19","slug":"%e7%94%a8lstm%e6%a8%a1%e5%9e%8b%e9%a2%84%e6%b5%8b%e6%af%94%e7%89%b9%e5%b8%81%e4%bb%b7%e6%a0%bc%ef%bc%9a%e5%87%86%e7%a1%ae%e7%8e%87%e6%8f%90%e5%8d%8730%e7%9a%84%e7%a7%98%e5%af%86","status":"publish","type":"post","link":"https:\/\/laoyulaoyu.com\/index.php\/2025\/02\/04\/%e7%94%a8lstm%e6%a8%a1%e5%9e%8b%e9%a2%84%e6%b5%8b%e6%af%94%e7%89%b9%e5%b8%81%e4%bb%b7%e6%a0%bc%ef%bc%9a%e5%87%86%e7%a1%ae%e7%8e%87%e6%8f%90%e5%8d%8730%e7%9a%84%e7%a7%98%e5%af%86\/","title":{"rendered":"\u7528LSTM\u6a21\u578b\u9884\u6d4b\u6bd4\u7279\u5e01\u4ef7\u683c\uff1a\u51c6\u786e\u7387\u63d0\u534730%\u7684\u79d8\u5bc6"},"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
\n
\u5199\u5728\u524d\u9762\u7684\u8bdd\uff1a<\/strong>\u6700\u8fd1\u7814\u7a76\u4e86\u4e00\u79cd\u57fa\u4e8eLSTM\u7684\u6bd4\u7279\u5e01\u4ef7\u683c\u9884\u6d4b\u6a21\u578b\uff0c\u901a\u8fc7\u4f18\u5316\u7b97\u6cd5\u548c\u7279\u5f81\u5de5\u7a0b\uff0c\u51c6\u786e\u7387\u63d0\u5347\u4e8630%\u3002\u8fd9\u7bc7\u6587\u7ae0\u5c06\u5e26\u4f60\u4e86\u89e3\u5982\u4f55\u5229\u7528\u6df1\u5ea6\u5b66\u4e60\u6280\u672f\uff0c\u4ece\u5386\u53f2\u6570\u636e\u4e2d\u6355\u6349\u6bd4\u7279\u5e01\u7684\u4ef7\u683c\u8d8b\u52bf<\/mark>\uff0c\u4e3a\u6295\u8d44\u51b3\u7b56\u63d0\u4f9b\u79d1\u5b66\u4f9d\u636e\u3002\u5982\u679c\u4f60\u5bf9\u6bd4\u7279\u5e01\u6295\u8d44\u611f\u5174\u8da3\uff0c\u8fd9\u7bc7\u5e72\u8d27\u4e0d\u5bb9\u9519\u8fc7\uff01<\/pre>\n<\/blockquote>\n\n\n\n
\u6bd4\u7279\u5e01\u4ef7\u683c\u6ce2\u52a8\u5267\u70c8\uff0c\u56e0\u6b64\u5bf9\u4ea4\u6613\u7b56\u7565\u800c\u8a00\uff0c\u5206\u949f\u7ea7\u7684\u7cbe\u51c6\u9884\u6d4b\u81f3\u5173\u91cd\u8981\u3002\u7136\u800c\uff0c\u76ee\u524d\u9488\u5bf9\u5206\u949f\u7ea7\u6bd4\u7279\u5e01\u4ef7\u683c\u9884\u6d4b\u7684\u7814\u7a76\u5e76\u4e0d\u591a\uff0c\u8fd9\u8ba9\u6211\u89c9\u5f97\u8fd9\u662f\u4e00\u4e2a\u503c\u5f97\u63a2\u7d22\u7684\u6709\u8da3\u8d77\u70b9\u3002<\/p>\n\n\n\n
\u4e00\u3001\u80cc\u666f\u8bf4\u660e<\/strong><\/h2>\n\n\n\n
\u6bd4\u7279\u5e01\uff08BTC\uff09\u662f\u4e00\u79cd\u53bb\u4e2d\u5fc3\u5316\u7684\u6570\u5b57\u8d27\u5e01\uff0c\u4e5f\u662f\u5168\u7403\u6700\u77e5\u540d\u7684\u52a0\u5bc6\u8d27\u5e01\u4e4b\u4e00\u3002\u4ece2018\u5e74\u5f00\u59cb\uff0c\u6bd4\u7279\u5e01\u7684\u4ef7\u683c\u7ecf\u5386\u4e86\u663e\u8457\u4e0a\u6da8\uff0c\u5438\u5f15\u4e86\u5927\u91cf\u6295\u8d44\u8005\u8fdb\u5165\u6570\u5b57\u8d27\u5e01\u5e02\u573a\u3002\u7136\u800c\uff0c\u52a0\u5bc6\u8d27\u5e01\u5e02\u573a\u7684\u9ad8\u5ea6\u6ce2\u52a8\u6027\u548c\u4e0d\u786e\u5b9a\u6027\u4e5f\u6fc0\u53d1\u4e86\u6211\u4eec\u5bf9\u4ef7\u683c\u9884\u6d4b\u7684\u6df1\u5165\u63a2\u7d22\uff0c\u5c24\u5176\u662f\u5728\u66f4\u7cbe\u7ec6\u7684\u65f6\u95f4\u5c3a\u5ea6\u4e0a\u8fdb\u884c\u5206\u6790\u548c\u9884\u6d4b\u3002<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\u5728\u8fc7\u53bb\u7684\u5341\u5e74\u91cc\uff0c\u6bd4\u7279\u5e01\u5728\u4ef7\u683c\u5f62\u6210\u3001\u6ce2\u52a8\u6027\u3001\u7cfb\u7edf\u52a8\u529b\u5b66\u548c\u7ecf\u6d4e\u4ef7\u503c\u7b49\u591a\u4e2a\u5c42\u9762\u5f97\u5230\u4e86\u5e7f\u6cdb\u7814\u7a76\u3002D. G. Baur \u7b49\u4eba\u7684\u7814\u7a76\u4ece\u7ecf\u6d4e\u5b66\u89d2\u5ea6\u63ed\u793a\u4e86\u6bd4\u7279\u5e01\u7684\u53cc\u91cd\u5c5e\u6027\uff1a\u65e2\u5177\u6709\u57fa\u672c\u4ef7\u503c\uff0c\u53c8\u5177\u5907\u6295\u673a\u4ef7\u503c\u3002\u4ed6\u4eec\u7684\u7814\u7a76\u8fdb\u4e00\u6b65\u8bc1\u5b9e\uff0c\u6bd4\u7279\u5e01\u662f\u4e00\u79cd\u9ad8\u5ea6\u52a8\u6001\u5316\u7684\u91d1\u878d\u8d44\u4ea7\uff0c\u5176\u4ef7\u683c\u6781\u6613\u53d7\u5230\u5e02\u573a\u6d3b\u52a8\u7684\u5f71\u54cd\u3002<\/p>\n\n\n\n
\u9274\u4e8e\u52a0\u5bc6\u8d27\u5e01\u4e0e\u80a1\u7968\u7b49\u4f20\u7edf\u6295\u8d44\u5de5\u5177\u5728\u5c5e\u6027\u4e0a\u7684\u76f8\u4f3c\u6027\uff0c\u52a0\u5bc6\u8d27\u5e01\u7684\u4ef7\u683c\u9884\u6d4b\u53ef\u4ee5\u501f\u9274\u80a1\u7968\u5e02\u573a\u7684\u9884\u6d4b\u65b9\u6cd5\u3002\u4f20\u7edf\u5206\u6790\u65b9\u6cd5\u548c\u673a\u5668\u5b66\u4e60\u65b9\u6cd5\u662f\u5e94\u5bf9\u80a1\u5e02\u6ce2\u52a8\u7684\u4e24\u79cd\u4e3b\u8981\u624b\u6bb5\u3002\u7136\u800c\uff0c\u4f20\u7edf\u5206\u6790\u65b9\u6cd5\u4f9d\u8d56\u4e8e\u8ba1\u91cf\u7ecf\u6d4e\u5b66\u6a21\u578b\u6216\u65b9\u7a0b\uff0c\u96be\u4ee5\u9002\u5e94\u9ad8\u5ea6\u52a8\u6001\u548c\u590d\u6742\u7684\u5e02\u573a\u73af\u5883\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u795e\u7ecf\u7f51\u7edc\u56e0\u5176\u5f3a\u5927\u7684\u52a8\u6001\u9884\u6d4b\u80fd\u529b\u8131\u9896\u800c\u51fa\uff0c\u80fd\u591f\u4ece\u6d77\u91cf\u6570\u636e\u4e2d\u81ea\u52a8\u63d0\u53d6\u6a21\u5f0f\uff0c\u800c\u65e0\u9700\u4f9d\u8d56\u5148\u9a8c\u77e5\u8bc6\u3002<\/p>\n\n\n\n
2018 \u5e74\u7684\u4e00\u9879\u7814\u7a76\u5bf9\u6bd4\u4e86 ARIMA \u65f6\u95f4\u5e8f\u5217\u6a21\u578b\u548c LSTM \u6df1\u5ea6\u5b66\u4e60\u7b97\u6cd5\u5728\u6bd4\u7279\u5e01\u4ef7\u683c\u9884\u6d4b\u4e2d\u7684\u8868\u73b0\u3002\u7ed3\u679c\u663e\u793a\uff0cARIMA \u6a21\u578b\u7684\u5e73\u5747\u7edd\u5bf9\u767e\u5206\u6bd4\u8bef\u5dee\uff08MAPE\uff09\u4e3a 11.86%\uff0c\u800c LSTM \u6a21\u578b\u7684 MAPE \u4ec5\u4e3a 1.40%\u3002\u7ed3\u5408\u5176\u4ed6\u7cbe\u5ea6\u6d4b\u8bd5\u7ed3\u679c\uff0c\u7814\u7a76\u8bc1\u5b9e LSTM \u5728\u65f6\u95f4\u5e8f\u5217\u4ef7\u683c\u9884\u6d4b\u4e2d\u8868\u73b0\u66f4\u4f18\u3002\u7136\u800c\uff0c\u6bcf\u65e5\u4ef7\u683c\u9884\u6d4b\u65b9\u6cd5\u5728\u5206\u949f\u7ea7\u6570\u636e\u4e0a\u7684\u8868\u73b0\u53ef\u80fd\u6709\u6240\u4e0d\u540c\uff0c\u56e0\u4e3a\u5feb\u901f\u7684\u4ef7\u683c\u6ce2\u52a8\u53ef\u80fd\u5bfc\u81f4\u4e25\u91cd\u7684\u8fc7\u62df\u5408\u95ee\u9898\u3002\u4e3a\u6b64\uff0c\u7814\u7a76\u8005\u63d0\u51fa\u4e86\u4e00\u79cd\u89e3\u51b3\u65b9\u6848\uff1a\u5728\u6bcf\u6b21\u8bad\u7ec3\u4e2d\u968f\u673a\u7701\u7565\u4e00\u534a\u7684\u7279\u5f81\u68c0\u6d4b\u5668\uff0c\u4ee5\u6709\u6548\u51cf\u5c11\u8fc7\u62df\u5408\u3002<\/p>\n\n\n\n
\u6b64\u5916\uff0c\u7814\u7a76\u8005\u8fd8\u63d0\u51fa\u4e86 LSTM \u7684\u6df7\u5408\u6a21\u578b\uff0c\u4e0e\u4f20\u7edf\u7684\u9012\u5f52\u795e\u7ecf\u7f51\u7edc\u76f8\u6bd4\uff0c\u8fd9\u79cd\u6df7\u5408\u6a21\u578b\u5177\u6709\u66f4\u6e05\u6670\u7684\u6570\u636e\u6d41\u548c\u7ed3\u6784\uff0c\u80fd\u591f\u66f4\u597d\u5730\u6355\u6349\u6570\u636e\u4e2d\u7684\u65f6\u53d8\u6548\u5e94\uff0c\u4ece\u800c\u8fdb\u4e00\u6b65\u63d0\u5347\u9884\u6d4b\u7cbe\u5ea6\u3002<\/p>\n\n\n\n
\u4e8c\u3001\u5e94\u7528\u6a21\u578b<\/strong><\/h2>\n\n\n\n
\u6211\u4eec\u7684\u76ee\u6807\u662f\u8bc4\u4f30\u6df7\u5408\u6a21\u578b\u4e0e LSTM \u5176\u4ed6\u53d8\u4f53\uff08\u5982 CNN-LSTM\u3001LSTM \u548c stacked-LSTM\uff09\u7684\u6027\u80fd\u5dee\u5f02\uff0c\u5e76\u63a2\u7d22\u4e00\u79cd\u6709\u6548\u7684\u6570\u636e\u9884\u5904\u7406\u65b9\u6cd5\uff0c\u4ee5\u63d0\u5347\u6df7\u5408\u6a21\u578b\u6216\u57fa\u7840\u6a21\u578b\u5728\u5904\u7406\u5206\u949f\u7ea7\u65f6\u95f4\u5e8f\u5217\u6570\u636e\u65f6\u7684\u8868\u73b0\u3002<\/p>\n\n\n\n
2.1 LSTM \u6a21\u578b<\/strong><\/h3>\n\n\n\n
LSTM\uff08\u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc\uff09\u7531 Hochreiter \u548c Schmidhuber \u4e8e 1997 \u5e74\u63d0\u51fa\uff0c\u65e8\u5728\u89e3\u51b3\u4f20\u7edf RNN\uff08\u9012\u5f52\u795e\u7ecf\u7f51\u7edc\uff09\u4e2d\u7684\u957f\u671f\u4f9d\u8d56\u95ee\u9898\u3002RNN \u901a\u8fc7\u9012\u5f52\u5c42\u7684\u5faa\u73af\u53cd\u9988\u4fdd\u5b58\u53c2\u6570\uff0c\u4f46\u7531\u4e8e\u4e0a\u4e0b\u6587\u4fe1\u606f\u7684\u5c40\u9650\u6027\uff0c\u5b83\u5728\u6355\u6349\u957f\u671f\u65f6\u95f4\u4f9d\u8d56\u6027\u65b9\u9762\u8868\u73b0\u6b20\u4f73\u3002\u6b64\u5916\uff0cRNN \u8fd8\u5bb9\u6613\u906d\u9047\u68af\u5ea6\u6d88\u5931\u95ee\u9898\u3002LSTM \u4f5c\u4e3a\u4e00\u79cd\u7279\u6b8a\u7684\u9012\u5f52\u795e\u7ecf\u7f51\u7edc\uff0c\u4e13\u95e8\u8bbe\u8ba1\u7528\u4e8e\u89e3\u51b3\u68af\u5ea6\u6d88\u5931\u95ee\u9898\uff0c\u5e76\u5728\u5e8f\u5217\u9884\u6d4b\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u8272\u3002LSTM \u7684\u6838\u5fc3\u5728\u4e8e\u5176\u5b58\u50a8\u5355\u5143\uff0c\u901a\u8fc7\u56db\u4e2a\u5173\u952e\u95e8\uff08\u9057\u5fd8\u95e8\u3001\u8f93\u5165\u95e8\u3001\u5355\u5143\u95e8\u548c\u8f93\u51fa\u95e8\uff09\u7684\u64cd\u4f5c\uff0c\u80fd\u591f\u957f\u65f6\u95f4\u4fdd\u5b58\u91cd\u8981\u53c2\u6570\u3002\u4e0b\u56fe\u5c55\u793a\u4e86\u6bcf\u4e2a LSTM \u5355\u5143\u7684\u8be6\u7ec6\u7ed3\u6784\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
LSTM \u6a21\u578b\u4ee3\u7801\u5982\u4e0b\uff1a<\/p>\n\n\n\n
class LSTMPredictor(nn.Module):\n    def __init__(self, sequence_dim, hidden_dim, drop_out_prob = 0.2):\n        super(LSTMPredictor, self).__init__()\n        self.sequence_dim = sequence_dim\n        self.hidden_dim = hidden_dim\n        self.lstm = nn.LSTM(sequence_dim, hidden_dim)\n        self.dropout = nn.Dropout(p=drop_out_prob)\n        self.lstm2 = nn.LSTM(sequence_dim, hidden_dim)\n        self.dropout = nn.Dropout(p=drop_out_prob)\n        self.lstm3 = nn.LSTM(sequence_dim, hidden_dim)\n        self.dropout = nn.Dropout(p=drop_out_prob)\n        self.hidden2target = nn.Linear(hidden_dim, 1)\n\n    def forward(self, X):\n        lstm_out, _ = self.lstm(X)\n        drop_out1 = self.dropout(lstm_out)\n\n        lstm_out2, _ = self.lstm(drop_out1)\n        drop_out2 = self.dropout(lstm_out2)\n\n        lstm_out3, _ = self.lstm(drop_out2)\n        drop_out3 = self.dropout(lstm_out3)\n\n        lstm_out4, _ = self.lstm(drop_out3)\n        drop_out = self.dropout(lstm_out4)\n\n        pred_out = self.hidden2target(drop_out)\n        return pred_out<\/code><\/pre>\n\n\n\n
2.2 \u5806\u53e0 LSTM \u6a21\u578b<\/strong><\/h3>\n\n\n\n
\u7531\u4e8e\u6bd4\u7279\u5e01\u9884\u6d4b\u6a21\u578b\u7684\u76ee\u6807\u662f\u5b66\u4e60\u65f6\u95f4\u5e8f\u5217\u6570\u636e\u7684\u591a\u5c42\u6b21\u7279\u5f81\uff0c\u56e0\u6b64\u9700\u8981\u5728\u6a21\u578b\u4e2d\u589e\u52a0\u66f4\u6df1\u7684\u9690\u85cf\u5c42\u4ee5\u63d0\u5347\u6027\u80fd\u3002\u901a\u8fc7\u5782\u76f4\u5806\u53e0 LSTM \u7ed3\u6784\uff0c\u6bcf\u4e00\u5c42 LSTM \u90fd\u4f1a\u8f93\u51fa\u4e00\u7ec4\u5411\u91cf\uff0c\u5e76\u4f5c\u4e3a\u4e0b\u4e00\u5c42 LSTM \u7684\u8f93\u5165\u3002\u8fd9\u79cd\u8bbe\u8ba1\u4f7f\u5f97\u6bcf\u4e00\u5c42\u7684\u9690\u85cf\u72b6\u6001\u80fd\u591f\u5728\u4e0d\u540c\u7684\u65f6\u95f4\u5c3a\u5ea6\u4e0a\u8fd0\u884c\uff0c\u4ece\u800c\u66f4\u597d\u5730\u6355\u6349\u6570\u636e\u7684\u590d\u6742\u6a21\u5f0f\u3002\u6b64\u5916\uff0c\u5728 LSTM \u5c42\u4e4b\u95f4\u8fd8\u52a0\u5165\u4e86\u5254\u9664\u5c42\uff08Dropout Layer\uff09\uff0c\u4ee5\u6709\u6548\u9632\u6b62\u6a21\u578b\u8fc7\u5ea6\u62df\u5408\uff0c\u786e\u4fdd\u5176\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n\n\n\n
2.3 CNN-LSTM \u6a21\u578b<\/strong><\/h3>\n\n\n\n
CNN-LSTM \u662f\u4e00\u79cd\u4e13\u95e8\u4e3a\u5e8f\u5217\u9884\u6d4b\u95ee\u9898\u8bbe\u8ba1\u7684\u6a21\u578b\uff0c\u5e7f\u6cdb\u5e94\u7528\u4e8e\u56fe\u50cf\u63cf\u8ff0\u3001\u89c6\u9891\u63cf\u8ff0\u548c\u6587\u672c\u63cf\u8ff0\u7b49\u4efb\u52a1\u3002\u5176\u6838\u5fc3\u7ed3\u6784\u7ed3\u5408\u4e86\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08CNN\uff09\u548c\u957f\u77ed\u671f\u8bb0\u5fc6\u7f51\u7edc\uff08LSTM\uff09\uff1a\u9996\u5148\u901a\u8fc7\u5377\u79ef\u5c42\u5bf9\u8f93\u5165\u6570\u636e\u8fdb\u884c\u7279\u5f81\u63d0\u53d6\uff0c\u7136\u540e\u5c06\u63d0\u53d6\u7684\u7279\u5f81\u4f20\u9012\u7ed9 LSTM \u5c42\u8fdb\u884c\u5e8f\u5217\u9884\u6d4b\u3002\u4ee5\u4e0b\u662f\u8be5\u6a21\u578b\u7684\u4ee3\u7801\u5b9e\u73b0\u6846\u67b6\uff1a<\/p>\n\n\n\n
import torch.nn as nn\nimport torch.nn.functional as F\n\nclass LSTMPredictor(nn.Module):\n    def __init__(self, sequence_dim, hidden_dim, drop_out_prob = 0.2):\n        super(LSTMPredictor, self).__init__()\n        self.sequence_dim = sequence_dim\n        self.hidden_dim = hidden_dim\n\n        self.conv1d = nn.Conv1d(1, 1, kernel_size=1)\n        self.lstm = nn.LSTM(sequence_dim, hidden_dim)\n        self.dropout = nn.Dropout(p=drop_out_prob)\n\n        self.hidden2target = nn.Linear(hidden_dim, 1)\n\n    def forward(self, X):\n        conv_out = self.conv1d(X)\n        lstm_out, _ = self.lstm(conv_out)\n        drop_out = self.dropout(lstm_out)\n\n        pred_out = self.hidden2target(drop_out)\n        return pred_out<\/code><\/pre>\n\n\n\n
\u5982\u56fe\u6240\u793a\uff0cCNN-LSTM \u9996\u5148\u4f7f\u7528\u4e86 nn.Conv1d(1,1,kernel_size=1)\u5c42\uff0c\u8fd9\u662f\u4e00\u4e2a\u4e00\u7ef4\u5377\u79ef\u5c42\uff0c\u5b83\u63a5\u6536\u5355\u901a\u9053\u8f93\u5165\uff0c\u5e94\u7528\u6838\u5927\u5c0f\u4e3a 1 \u7684\u5377\u79ef\uff0c\u5e76\u4ea7\u751f\u5355\u901a\u9053\u8f93\u51fa\u3002<\/p>\n\n\n\n
\u4e09\u3001\u9879\u76ee\u5b9e\u8df5<\/strong><\/h2>\n\n\n\n

3.1 \u6570\u636e<\/strong><\/h3>\n\n\n\n
\u672c\u9879\u76ee\u4f7f\u7528\u7684\u6570\u636e\u6db5\u76d6\u4e86 2019 \u5e74 4 \u6708 1 \u65e5\u81f3 2019 \u5e74 3 \u6708 2 \u65e5\u671f\u95f4\u4ee5\u7f8e\u5143\u8ba1\u4ef7\u7684\u6bd4\u7279\u5e01\uff08BTC\uff09\u5206\u949f\u7ea7\u5386\u53f2\u6570\u636e\uff0c\u6bd4\u7279\u5e01\u662f\u5e02\u503c\u6700\u9ad8\u7684\u52a0\u5bc6\u8d27\u5e01\u3002\u6570\u636e\u6765\u6e90\u4e8e www.kaggle.com\u3002<\/a>\u4e3a\u4e86\u4fbf\u4e8e\u6a21\u578b\u8bc4\u4f30\uff0c\u6570\u636e\u96c6\u88ab\u5212\u5206\u4e3a\u8bad\u7ec3\u96c6\u3001\u9a8c\u8bc1\u96c6\u548c\u6d4b\u8bd5\u96c6\u4e09\u90e8\u5206\u3002\u5177\u4f53\u5212\u5206\u5982\u4e0b\uff1a\u8bad\u7ec3\u96c6\u5305\u542b 2019 \u5e74 4 \u6708 1 \u65e5\u81f3 2019 \u5e74 4 \u6708 25 \u65e5\u7684\u5206\u949f\u7ea7\u6570\u636e\uff08\u5171 26,394 \u4e2a\u6570\u636e\u70b9\uff09\uff0c\u9a8c\u8bc1\u96c6\u5305\u542b 2019 \u5e74 4 \u6708 26 \u65e5\u81f3 2019 \u5e74 4 \u6708 29 \u65e5\u7684\u5206\u949f\u7ea7\u6570\u636e\uff08\u5171 6,440 \u4e2a\u6570\u636e\u70b9\uff09\uff0c\u6d4b\u8bd5\u96c6\u5219\u5305\u542b 2019 \u5e74 4 \u6708 30 \u65e5\u81f3 2019 \u5e74 3 \u6708 2 \u65e5\u7684\u5206\u949f\u7ea7\u6570\u636e\uff08\u5171 6,440 \u4e2a\u6570\u636e\u70b9\uff09\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u4e0a\u56fe\u663e\u793a\uff0c\u5728\u6570\u636e\u6536\u96c6\u671f\u95f4\uff0c\u5206\u949f\u7ea7\u522b\u7684 BTC \u503c\u6ce2\u52a8\u5f88\u5927\u3002\u8fd9\u4e9b\u6ce2\u52a8\u5728 2019 \u5e74 4 \u6708 26 \u65e5\u81f3 4 \u6708 28 \u65e5\u671f\u95f4\u6700\u4e3a\u7a81\u51fa\u3002\u4e00\u4e2a\u89c2\u5bdf\u7ed3\u679c\u662f\uff0c\u6bd4\u7279\u5e01\u5e02\u573a\u5728\u6b64\u671f\u95f4\u51fa\u73b0\u4e86\u76f8\u5f53\u5927\u7684\u6ce2\u52a8\uff0c\u504f\u79bb\u4e86\u6b63\u5e38\u80a1\u7968\u5e02\u573a\u7684\u5e38\u89c4\u884c\u4e3a\u3002\u8fd9\u4e00\u65f6\u671f\u7684\u6570\u636e\u867d\u7136\u5df2\u7ecf\u8fc7\u65f6\uff0c\u4f46\u4ecd\u7136\u5f88\u597d\u5730\u4f53\u73b0\u4e86\u6bd4\u7279\u5e01\u5e02\u573a\u7684\u4e0d\u786e\u5b9a\u6027\u89c4\u5f8b\u6027\uff0c\u8fd9\u4f7f\u5f97\u4f20\u7edf\u7684\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u6a21\u578b\u4e0d\u592a\u5bb9\u6613\u6293\u4f4f\u6570\u636e\u7684\u672c\u8d28\u3002<\/p>\n\n\n\n
\u4e3a\u5168\u9762\u4e86\u89e3\u6570\u636e\uff0c\u6211\u4eec\u5728\u8868 1 \u4e2d\u5217\u51fa\u4e86\u5bf9\u6211\u4eec\u7684\u6570\u636e\u96c6\u8fdb\u884c\u7684\u589e\u5f3a\u578b Dickey-Fuller \uff08ADF\uff09\u68c0\u9a8c\u7684 t \u7edf\u8ba1\u91cf\u548c\u76f8\u5173 p \u503c\u7684\u7ed3\u679c\u3002\u6211\u4eec\u9009\u62e9\u4e86 5%\u4e34\u754c\u6c34\u5e73\u7684\u7edf\u8ba1\u663e\u8457\u6027\uff08\u89c1\u8868 1\uff09\u3002\u8868 1 \u7684\u89e3\u91ca\u8868\u660e\uff0cBTC \u65f6\u95f4\u5e8f\u5217\u610f\u5473\u7740\u8fd9\u4e9b\u5e8f\u5217\u662f\u975e\u5e73\u7a33\u7684\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n

3.2 \u6570\u636e\u9884\u5904\u7406<\/strong><\/h3>\n\n\n\n

\u4e3a\u4e86\u51c6\u5907\u8bad\u7ec3\u6570\u636e\uff0c\u6211\u4eec\u5bf9\u539f\u59cb\u6570\u636e\u96c6\u8fdb\u884c\u4e86\u4e24\u9879\u5173\u952e\u5904\u7406\u3002\u9996\u5148\uff0c\u5bf9\u6570\u636e\u96c6\u8fdb\u884c\u4e86\u5f52\u4e00\u5316\u5904\u7406\uff0c\u76ee\u7684\u662f\u5728\u4e0d\u626d\u66f2\u6570\u503c\u8303\u56f4\u5dee\u5f02\u7684\u60c5\u51b5\u4e0b\uff0c\u5c06\u6570\u503c\u5217\u7684\u6570\u503c\u8f6c\u6362\u4e3a\u7edf\u4e00\u5c3a\u5ea6\u3002\u5176\u6b21\uff0c\u91c7\u7528\u4e86\u6ed1\u52a8\u7a97\u53e3\u7b97\u6cd5\uff0c\u8fd9\u79cd\u65b9\u6cd5\u7279\u522b\u9002\u7528\u4e8e\u9884\u6d4b\u9ad8\u5ea6\u975e\u7ebf\u6027\u7684\u80a1\u7968\u6570\u636e\u3002<\/p>\n\n\n\n
\u5728\u5f52\u4e00\u5316\u5904\u7406\u4e2d\uff0c\u6211\u4eec\u4f7f\u7528\u4e86\u96f6\u57fa\u5f52\u4e00\u5316\uff08Zero-based Normalization\uff09\u3002\u96f6\u57fa\u5f52\u4e00\u5316\u7684\u6838\u5fc3\u601d\u60f3\u662f\u53cd\u6620\u6570\u636e\u76f8\u5bf9\u4e8e\u7b2c\u4e00\u4e2a\u6761\u76ee\u7684\u53d8\u5316\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u7ed9\u5b9a\u4e00\u4e2a\u5927\u5c0f\u4e3a n \u7684\u5411\u91cf V\uff0c\u901a\u8fc7\u4ee5\u4e0b\u516c\u5f0f\u6784\u5efa\u65b0\u7684\u5f52\u4e00\u5316\u5411\u91cf V’\uff1a<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u751f\u6210 V\u2032\u540e\uff0c\u8bb0\u5f55\u6240\u6709\u8f6c\u6362 V\u2032\u5411\u91cf\u4e2d\u7684\u6700\u5c0f\u503c Vmin \u548c\u6700\u5927\u503c Vmax\u3002\u7136\u540e\uff0c\u901a\u8fc7\u4ee5\u4e0b\u65b9\u6cd5\u6784\u5efa\u51fa\u6700\u7ec8\u7ed3\u679c V\u2032\u2032\uff1a<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u7ecf\u8fc7\u8fd9\u6837\u7684\u5904\u7406\uff0c\u539f\u59cb\u6570\u636e\u96c6\u4e2d\u7684\u6570\u636e\u987a\u5e8f\u5f97\u4ee5\u4fdd\u7559\uff0c\u4ece\u800c\u4f7f\u795e\u7ecf\u7f51\u7edc\u66f4\u5bb9\u6613\u8bad\u7ec3\u3002<\/p>\n\n\n\n
\u9664\u4e86\u5f52\u4e00\u5316\u5904\u7406\uff0c\u6211\u4eec\u8fd8\u5f15\u5165\u4e86\u6ed1\u52a8\u7a97\u53e3\u7b97\u6cd5\uff0c\u8fd9\u662f\u4e00\u79cd\u5e7f\u6cdb\u5e94\u7528\u4e8e\u65f6\u95f4\u5e8f\u5217\u6570\u636e\u5904\u7406\u7684\u6280\u672f\u3002\u8be5\u7b97\u6cd5\u7684\u6838\u5fc3\u601d\u60f3\u662f\u9009\u62e9\u4e00\u4e2a\u7279\u5b9a\u65f6\u95f4\u70b9 tt<\/em> \u4ee5\u53ca\u4e00\u4e2a\u56fa\u5b9a\u7684\u7a97\u53e3\u957f\u5ea6 win_lenwin<\/em>_len<\/em>\uff0c\u7136\u540e\u4ece\u539f\u59cb\u6570\u636e\u96c6\u7684 t+win_len+1t<\/em>+win<\/em>_len<\/em>+1 \u4f4d\u7f6e\u63d0\u53d6\u5bf9\u5e94\u7684\u6807\u7b7e yy<\/em>\uff08\u5982\u4e0b\u56fe\u6240\u793a\uff09\u3002\u6ed1\u52a8\u7a97\u53e3\u7b97\u6cd5\u80fd\u591f\u5145\u5206\u5229\u7528\u65f6\u95f4\u5e8f\u5217\u6570\u636e\u7684\u8fde\u7eed\u6027\uff0c\u540c\u65f6\u4fdd\u6301\u6570\u636e\u7684\u5185\u5728\u5e8f\u5217\u7ed3\u6784\u3002\u5728\u672c\u9879\u76ee\u4e2d\uff0c\u6211\u4eec\u9009\u62e9\u7684\u7a97\u53e3\u957f\u5ea6\u4e3a win_len=5win<\/em>_len<\/em>=5\u3002\u56e0\u6b64\uff0c\u7ecf\u8fc7\u5904\u7406\u540e\uff0c\u8bad\u7ec3\u8f93\u5165\u6570\u636e\u7684\u7ef4\u5ea6\u4e3a (x,5,5)(x<\/em>,5,5)\uff0c\u5176\u4e2d xx<\/em> \u8868\u793a\u6570\u636e\u96c6\u7684\u5927\u5c0f\uff0c\u7b2c\u4e00\u4e2a 5 \u4ee3\u8868\u6279\u91cf\u5927\u5c0f batchbatch<\/em>\uff0c\u7b2c\u4e8c\u4e2a 5 \u5219\u662f\u6240\u9009\u7684\u7a97\u53e3\u957f\u5ea6 win_lenwin<\/em>_len<\/em>\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n

3.3 \u53c2\u6570<\/strong><\/h3>\n\n\n\n
\u8d85\u53c2\u6570\u901a\u8fc7\u7f51\u683c\u641c\u7d22\u6280\u672f\u8fdb\u884c\u8c03\u6574\u3002\u4e0b\u8868\u6c47\u603b\u4e86\u4e3a\u6bcf\u4e2a\u6a21\u578b\u9009\u62e9\u7684\u53c2\u6570\uff0c\u4ee5\u786e\u4fdd\u53ef\u91cd\u590d\u6027\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n

3.4 \u65b9\u6cd5<\/strong><\/h3>\n\n\n\n
\u4e3a\u4e86\u8bc4\u4f30\u6027\u80fd\uff0c\u91c7\u7528\u4e86\u5747\u65b9\u8bef\u5dee (MSE)\u3001\u5747\u65b9\u6839\u8bef\u5dee (RMSE)\u3001\u5e73\u5747\u975e\u6eb6\u8d28\u8bef\u5dee (MAE) \u548c\u5e73\u5747\u7edd\u5bf9\u767e\u5206\u6bd4\u8bef\u5dee (MAPE) \u4f5c\u4e3a\u6027\u80fd\u6307\u6807\u3002\u8ba1\u7b97\u516c\u5f0f\u5982\u4e0b<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u5176\u4e2d\uff0cP \u662f\u6bd4\u7279\u5e01\u7684\u5b9e\u9645\u4ef7\u683c\uff0cP\u2032 \u662f\u6bd4\u7279\u5e01\u7684\u9884\u6d4b\u4ef7\u683c\u3002<\/p>\n\n\n\n
3.5 \u7ed3\u679c<\/strong><\/h3>\n\n\n\n
\u5904\u7406\u540e\u7684\u8bad\u7ec3\u6570\u636e\u96c6\u88ab\u5206\u522b\u7528\u4e8e\u8bad\u7ec3 LSTM\u3001Stacked LSTM \u548c CNN-LSTM \u6a21\u578b\u3002\u8bad\u7ec3\u5b8c\u6210\u540e\uff0c\u6211\u4eec\u5229\u7528\u8fd9\u4e9b\u6a21\u578b\u5bf9\u6d4b\u8bd5\u6570\u636e\u96c6\u8fdb\u884c\u9884\u6d4b\uff0c\u5e76\u5c06\u9884\u6d4b\u7ed3\u679c\u4e0e\u5b9e\u9645\u503c\u8fdb\u884c\u6bd4\u8f83\u3002\u4e0b\u56fe\u5c55\u793a\u4e86 LSTM\u3001Stacked LSTM \u548c CNN-LSTM \u6a21\u578b\u7684\u9884\u6d4b\u503c\u4e0e\u5b9e\u9645\u503c\u7684\u5bf9\u6bd4\u60c5\u51b5\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n
\u7ecf\u8fc7\u591a\u6b21\u5b9e\u9a8c\uff0c\u4e3a\u9a8c\u8bc1\u6570\u636e\u96c6\u4e0a\u7684\u6bcf\u4e2a\u6a21\u578b\u627e\u5230\u4e86\u5408\u9002\u7684\u8d85\u53c2\u6570\u548c\u6a21\u578b\u53c2\u6570\u3002\u6bcf\u4e2a\u6a21\u578b\u90fd\u8fd0\u884c\u591a\u6b21\uff0c\u4ee5\u83b7\u5f97\u4e0d\u540c\u4f30\u8ba1\u6307\u6807\u7684\u5e73\u5747\u503c\u3002\u7ed3\u679c\u5982\u4e0b\u8868\u6240\u793a\u3002<\/p>\n\n\n\n
\"\"\/<\/figure>\n\n\n\n

3.6 \u5206\u6790<\/strong><\/h3>\n\n\n\n
\u4ece\u4e0a\u8868\u53ef\u4ee5\u770b\u51fa\uff0cCNN-LSTM \u7684 MSE\uff08\u5747\u65b9\u8bef\u5dee\uff09\u3001RMSE\uff08\u5747\u65b9\u6839\u8bef\u5dee\uff09\u548c MAE\uff08\u5e73\u5747\u7edd\u5bf9\u8bef\u5dee\uff09\u5728\u4e09\u4e2a\u6a21\u578b\u4e2d\u5747\u8868\u73b0\u6700\u4f73\u3002\u4e0e\u57fa\u7ebf LSTM \u76f8\u6bd4\uff0c\u5806\u53e0 LSTM \u7684 MAE \u4ece 27.89 \u964d\u81f3 25.86\uff0cRMSE \u4ece 52.83 \u964d\u81f3 50.85\u3002\u8fd9\u4e00\u6027\u80fd\u63d0\u5347\u7b26\u5408\u9884\u671f\uff0c\u56e0\u4e3a\u5806\u53e0 LSTM \u7684\u6bcf\u4e00\u5c42\u90fd\u52a0\u5165\u4e86\u5254\u9664\u5c42\uff08Dropout Layer\uff09\uff0c\u6709\u6548\u9632\u6b62\u4e86\u6a21\u578b\u8fc7\u62df\u5408\uff0c\u4ece\u800c\u51cf\u5c11\u4e86\u6570\u636e\u632f\u8361\u7684\u53ef\u80fd\u6027\u3002<\/p>\n\n\n\n
\u6b64\u5916\uff0cCNN-LSTM \u7684\u8868\u73b0\u5c24\u4e3a\u7a81\u51fa\uff0c\u5176 MSE\u3001RMSE \u548c MAE \u5206\u522b\u4e3a 7.041\u300126.53 \u548c 22.41\uff0c\u5747\u4e3a\u4e09\u4e2a\u6a21\u578b\u4e2d\u7684\u6700\u4f4e\u503c\u3002\u8fd9\u79cd\u663e\u8457\u7684\u6027\u80fd\u63d0\u5347\u4e3b\u8981\u5f52\u529f\u4e8e\u65b0\u589e\u7684 CNN \u5c42\u3002CNN \u5c42\u6210\u529f\u5730\u4ece\u6570\u636e\u4e2d\u63d0\u53d6\u5e76\u5206\u7c7b\u91cd\u8981\u7279\u5f81\uff0c\u4f7f\u6a21\u578b\u5bf9\u539f\u59cb\u6570\u636e\u96c6\u7684\u5fae\u5c0f\u53d8\u5316\u66f4\u52a0\u9c81\u68d2\uff0c\u4ece\u800c\u63d0\u9ad8\u4e86\u9884\u6d4b\u7cbe\u5ea6\u3002<\/p>\n\n\n\n
\u56db\u3001\u89c2\u70b9\u603b\u7ed3<\/strong><\/h2>\n\n\n\n
\u5728\u8fd9\u7bc7\u6587\u7ae0\u4e2d\uff0c\u6211\u4eec\u63a2\u7d22\u4e86\u4e09\u79cd\u57fa\u4e8e\u524d\u51e0\u5206\u949f\u6536\u76d8\u4ef7\u9884\u6d4b\u4e0b\u4e00\u5206\u949f\u52a0\u5bc6\u8d27\u5e01\u6536\u76d8\u4ef7\u7684\u6a21\u578b\uff1aLSTM\u3001\u5806\u53e0 LSTM \u548c CNN-LSTM\u3002\u7814\u7a76\u7ed3\u679c\u8868\u660e\uff0cCNN-LSTM \u6a21\u578b\u5728\u9884\u6d4b\u6bd4\u7279\u5e01\u5e02\u573a\u5206\u949f\u7ea7\u6536\u76d8\u4ef7\u65b9\u9762\u8868\u73b0\u6700\u4f73\u3002\u901a\u8fc7\u5728\u6bd4\u7279\u5e01\u5e02\u573a\u4e2d\u7684\u6d4b\u8bd5\uff0c\u6211\u4eec\u53d1\u73b0 CNN-LSTM \u6a21\u578b\u7684\u9884\u6d4b\u6027\u80fd\u663e\u8457\u4f18\u4e8e\u539f\u59cb LSTM \u6a21\u578b\uff0c\u5176 MSE\uff08\u5747\u65b9\u8bef\u5dee\uff09\u548c MAE\uff08\u5e73\u5747\u7edd\u5bf9\u8bef\u5dee\uff09\u5f97\u5206\u5728\u7edf\u8ba1\u4e0a\u5177\u6709\u660e\u663e\u4f18\u52bf\u3002<\/p>\n\n\n\n