基于LSTM的股票未来60天预测异常问题求助

问题诊断 #

当前代码仅完成了历史验证区间的价格拟合，未实现未来时序的迭代预测逻辑；同时训练轮次仅1轮，模型拟合程度极低，导致预测效果差。要实现2024年6月后的股价预测，需补充未来预测的核心循环逻辑，并优化模型训练流程。

解决方案步骤 #

1. 优化模型训练，提升拟合效果 #

增加训练轮次，加入早停和模型保存回调，避免过拟合同时保证模型收敛：

# 替换原训练代码部分
checkpoint = ModelCheckpoint('best_model.h5', monitor='loss', save_best_only=True)
early_stop = EarlyStopping(monitor='loss', patience=5, restore_best_weights=True)

model.fit(x_train, y_train, batch_size=32, epochs=20, callbacks=[checkpoint, early_stop])

2. 实现未来60天预测的核心逻辑 #

基于最新的60天历史数据，迭代预测每一天的价格，并将预测值作为下一天的输入特征，循环生成未来60天的预测结果：

# 获取最新60天的缩放后数据
last_60_days = scaled_data[-60:]
future_predictions = []

for _ in range(60):
    # 重塑数据为模型输入格式
    X_pred = last_60_days.reshape(1, 60, 1)
    # 预测当日价格
    pred_price = model.predict(X_pred, verbose=0)
    # 将预测值加入列表
    future_predictions.append(pred_price[0][0])
    # 更新last_60_days，移除最早一天，加入最新预测值
    last_60_days = np.append(last_60_days[1:], pred_price, axis=0)

# 反缩放，还原真实价格
future_predictions = scaler.inverse_transform(np.array(future_predictions).reshape(-1,1))

3. 生成未来日期序列并可视化 #

创建未来60天的日期索引，将预测结果与历史数据合并后绘图：

# 生成未来60天日期
last_date = data.index[-1]
future_dates = pd.date_range(start=last_date + pd.Timedelta(days=1), periods=60)

# 创建预测结果DataFrame
future_df = pd.DataFrame(future_predictions, index=future_dates, columns=['Future Predictions'])

# 合并历史数据与预测数据
combined_data = pd.concat([data, future_df])

# 可视化
plt.figure(figsize=(16,8))
plt.title('腾讯股价历史走势与未来60天预测')
plt.xlabel('日期', fontsize=18)
plt.ylabel('收盘价', fontsize=18)
plt.plot(combined_data['Close'], label='历史收盘价')
plt.plot(combined_data['Future Predictions'], label='未来60天预测', linestyle='--', color='red')
plt.legend(loc='lower right')
plt.xticks(rotation=45)
plt.show()

修改后的完整代码 #

import math
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
from mplfinance.original_flavor import candlestick_ohlc
from sklearn.preprocessing import MinMaxScaler
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import LSTM, Dense
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping

# 假设Tencent2是已加载的数据集
# Tencent2 = pd.read_csv('your_data.csv')

# 数据预处理
Tencent2['Date'] = pd.to_datetime(Tencent2['Date'])
Tencent2.set_index('Date', inplace=True)
data = Tencent2.filter(['Close'])
dataset = data.values
training_data_len = math.ceil(len(dataset) * 0.7)

# 数据缩放
scaler = MinMaxScaler(feature_range=(0,1))
scaled_data = scaler.fit_transform(dataset)

# 构建训练集
train_data = scaled_data[0:training_data_len, :]
x_train, y_train = [], []
for i in range(60, len(train_data)):
    x_train.append(train_data[i-60:i, 0])
    y_train.append(train_data[i, 0])

x_train, y_train = np.array(x_train), np.array(y_train)
x_train = np.reshape(x_train, (x_train.shape[0], x_train.shape[1], 1))

# 构建LSTM模型
model = Sequential()
model.add(LSTM(50, return_sequences=True, input_shape=(x_train.shape[1], 1)))
model.add(LSTM(50, return_sequences=False))
model.add(Dense(25))
model.add(Dense(1))

model.compile(optimizer='adam', loss='mean_squared_error')

# 训练模型
checkpoint = ModelCheckpoint('best_model.h5', monitor='loss', save_best_only=True)
early_stop = EarlyStopping(monitor='loss', patience=5, restore_best_weights=True)
model.fit(x_train, y_train, batch_size=32, epochs=20, callbacks=[checkpoint, early_stop])

# 历史验证（可选）
test_data = scaled_data[training_data_len - 60:]
X_test = []
y_test = dataset[training_data_len:, :]
for i in range(60, len(test_data)):
    X_test.append(test_data[i-60:i, 0])

X_test = np.array(X_test)
X_test = np.reshape(X_test, (X_test.shape[0], X_test.shape[1], 1))
predictions = model.predict(X_test)
predictions = scaler.inverse_transform(predictions)

# 未来60天预测
last_60_days = scaled_data[-60:]
future_predictions = []

for _ in range(60):
    X_pred = last_60_days.reshape(1, 60, 1)
    pred_price = model.predict(X_pred, verbose=0)
    future_predictions.append(pred_price[0][0])
    last_60_days = np.append(last_60_days[1:], pred_price, axis=0)

future_predictions = scaler.inverse_transform(np.array(future_predictions).reshape(-1,1))

# 生成未来日期
last_date = data.index[-1]
future_dates = pd.date_range(start=last_date + pd.Timedelta(days=1), periods=60)
future_df = pd.DataFrame(future_predictions, index=future_dates, columns=['Future Predictions'])

# 合并数据并可视化
combined_data = pd.concat([data, future_df])
plt.figure(figsize=(16,8))
plt.title('腾讯股价历史走势与未来60天预测')
plt.xlabel('日期', fontsize=18)
plt.ylabel('收盘价', fontsize=18)
plt.plot(combined_data['Close'], label='历史收盘价')
plt.plot(combined_data['Future Predictions'], label='未来60天预测', linestyle='--', color='red')
plt.legend(loc='lower right')
plt.xticks(rotation=45)
plt.show()

内容的提问来源于stack exchange，提问作者Spatdy