📘 基于长短期记忆网络（LSTM）的黄金价格预测模型/ALL.ipynb

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import plotly.express as px
from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_absolute_percentage_error, mean_squared_error, mean_absolute_error
import tensorflow as tf
from keras import Model
from keras.layers import Input, Dense, Dropout, LSTM
from keras.callbacks import EarlyStopping

# 忽略警告
import warnings
warnings.filterwarnings('ignore')

# 数据加载与基本检查
df = pd.read_csv('Gold Price (2013-2022).csv')
df['Date'] = pd.to_datetime(df['Date'])
df.sort_values(by='Date', ascending=True, inplace=True)
df.reset_index(drop=True, inplace=True)

# 删除不必要的特征
df.drop(['Vol.', 'Change %'], axis=1, inplace=True)

# 去除逗号并转换数据类型
num_cols = df.columns.drop(['Date'])
df[num_cols] = df[num_cols].replace({',': ''}, regex=True).astype('float64')

# 数据可视化
fig = px.line(y=df.Price, x=df.Date, title="Gold Price History Data")
fig.update_traces(line_color='black')
fig.update_layout(
    xaxis_title="Date", yaxis_title="Price",
    plot_bgcolor='rgba(255,223,0,0.8)'
)
fig.show()

# 数据分割：训练集和测试集
test_size = df[df.Date.dt.year == 2022].shape[0]
train_data = df.Price[:-test_size]
test_data = df.Price[-test_size-60:]  # 测试集包含滑动窗口部分

# 数据归一化
scaler = MinMaxScaler()
train_scaled = scaler.fit_transform(train_data.values.reshape(-1, 1))
test_scaled = scaler.transform(test_data.values.reshape(-1, 1))

# 滑动窗口构建数据
window_size = 60

# 训练集
X_train, y_train = [], []
for i in range(window_size, len(train_scaled)):
    X_train.append(train_scaled[i-window_size:i, 0])
    y_train.append(train_scaled[i, 0])

# 测试集
X_test, y_test = [], []
for i in range(window_size, len(test_scaled)):
    X_test.append(test_scaled[i-window_size:i, 0])
    y_test.append(test_scaled[i, 0])

# 转换为Numpy数组
X_train, y_train = np.array(X_train), np.array(y_train)
X_test, y_test = np.array(X_test), np.array(y_test)

# 调整输入形状
X_train = np.reshape(X_train, (X_train.shape[0], X_train.shape[1], 1))
X_test = np.reshape(X_test, (X_test.shape[0], X_test.shape[1], 1))

# 定义优化的LSTM模型
def define_optimized_model():
    input_layer = Input(shape=(window_size, 1))
    x = LSTM(64, return_sequences=True, activation='tanh')(input_layer)
    x = Dropout(0.2)(x)
    x = LSTM(32, return_sequences=False, activation='tanh')(x)
    x = Dropout(0.2)(x)
    output_layer = Dense(1, activation='linear')(x)

    model = Model(inputs=input_layer, outputs=output_layer)
    model.compile(loss='mean_squared_error', optimizer='adam')
    return model

# 模型训练
model = define_optimized_model()
early_stop = EarlyStopping(monitor='val_loss', patience=10, restore_best_weights=True)

history = model.fit(
    X_train, y_train, 
    epochs=100, batch_size=32, 
    validation_split=0.2, verbose=1,
    callbacks=[early_stop]
)

# 模型评估
y_pred_scaled = model.predict(X_test)
y_test_true = scaler.inverse_transform(y_test.reshape(-1, 1))
y_test_pred = scaler.inverse_transform(y_pred_scaled)

# 计算评估指标
mse = mean_squared_error(y_test_true, y_test_pred)
mae = mean_absolute_error(y_test_true, y_test_pred)
mape = mean_absolute_percentage_error(y_test_true, y_test_pred)
r2 = 1 - (np.sum((y_test_true - y_test_pred) ** 2) / np.sum((y_test_true - np.mean(y_test_true)) ** 2))

print(f"Test MSE: {mse:.4f}")
print(f"Test MAE: {mae:.4f}")
print(f"Test MAPE: {mape:.2%}")
print(f"Test R²: {r2:.4f}")

# 可视化模型预测效果
plt.figure(figsize=(15, 6), dpi=150)
plt.plot(df['Date'].iloc[:-test_size], scaler.inverse_transform(train_scaled), color='black', label='Training Data')
plt.plot(df['Date'].iloc[-test_size:], y_test_true, color='blue', label='Actual Test Data')
plt.plot(df['Date'].iloc[-test_size:], y_test_pred, color='red', label='Predicted Test Data')
plt.title('Model Performance on Gold Price Prediction', fontsize=15)
plt.xlabel('Date', fontsize=12)
plt.ylabel('Price', fontsize=12)
plt.legend(loc='upper left', prop={'size': 12})
plt.grid(color='gray', linestyle='--')
plt.show()

Epoch 1/100
57/57 [==============================] - 4s 34ms/step - loss: 0.0082 - val_loss: 0.0027
Epoch 2/100
57/57 [==============================] - 1s 25ms/step - loss: 0.0018 - val_loss: 0.0041
Epoch 3/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0016 - val_loss: 0.0015
Epoch 4/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0014 - val_loss: 0.0017
Epoch 5/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0016 - val_loss: 0.0045
Epoch 6/100
57/57 [==============================] - 1s 26ms/step - loss: 0.0012 - val_loss: 0.0044
Epoch 7/100
57/57 [==============================] - 1s 25ms/step - loss: 0.0012 - val_loss: 0.0028
Epoch 8/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0011 - val_loss: 0.0019
Epoch 9/100
57/57 [==============================] - 1s 25ms/step - loss: 0.0012 - val_loss: 0.0017
Epoch 10/100
57/57 [==============================] - 1s 26ms/step - loss: 0.0011 - val_loss: 9.5650e-04
Epoch 11/100
57/57 [==============================] - 2s 28ms/step - loss: 0.0011 - val_loss: 8.8959e-04
Epoch 12/100
57/57 [==============================] - 2s 28ms/step - loss: 0.0011 - val_loss: 8.6934e-04
Epoch 13/100
57/57 [==============================] - 2s 28ms/step - loss: 9.3384e-04 - val_loss: 0.0015
Epoch 14/100
57/57 [==============================] - 2s 29ms/step - loss: 9.5245e-04 - val_loss: 0.0010
Epoch 15/100
57/57 [==============================] - 2s 28ms/step - loss: 8.8055e-04 - val_loss: 7.6218e-04
Epoch 16/100
57/57 [==============================] - 2s 29ms/step - loss: 9.5084e-04 - val_loss: 0.0011
Epoch 17/100
57/57 [==============================] - 2s 29ms/step - loss: 8.6539e-04 - val_loss: 0.0012
Epoch 18/100
57/57 [==============================] - 2s 29ms/step - loss: 8.0801e-04 - val_loss: 0.0012
Epoch 19/100
57/57 [==============================] - 2s 29ms/step - loss: 7.3625e-04 - val_loss: 7.2624e-04
Epoch 20/100
57/57 [==============================] - 2s 28ms/step - loss: 7.4719e-04 - val_loss: 7.2584e-04
Epoch 21/100
57/57 [==============================] - 2s 29ms/step - loss: 7.7573e-04 - val_loss: 9.8627e-04
Epoch 22/100
57/57 [==============================] - 2s 28ms/step - loss: 6.2333e-04 - val_loss: 9.3673e-04
Epoch 23/100
57/57 [==============================] - 2s 29ms/step - loss: 6.5290e-04 - val_loss: 0.0025
Epoch 24/100
57/57 [==============================] - 2s 29ms/step - loss: 6.5372e-04 - val_loss: 0.0042
Epoch 25/100
57/57 [==============================] - 2s 29ms/step - loss: 6.5522e-04 - val_loss: 0.0017
Epoch 26/100
57/57 [==============================] - 2s 28ms/step - loss: 6.8554e-04 - val_loss: 0.0030
Epoch 27/100
57/57 [==============================] - 2s 28ms/step - loss: 5.9007e-04 - val_loss: 0.0011
Epoch 28/100
57/57 [==============================] - 2s 28ms/step - loss: 5.8071e-04 - val_loss: 9.5184e-04
Epoch 29/100
57/57 [==============================] - 2s 28ms/step - loss: 6.0606e-04 - val_loss: 5.8515e-04
Epoch 30/100
57/57 [==============================] - 2s 28ms/step - loss: 5.5738e-04 - val_loss: 5.8165e-04
Epoch 31/100
57/57 [==============================] - 2s 28ms/step - loss: 5.3708e-04 - val_loss: 5.5287e-04
Epoch 32/100
57/57 [==============================] - 2s 28ms/step - loss: 4.7901e-04 - val_loss: 5.2770e-04
Epoch 33/100
57/57 [==============================] - 2s 28ms/step - loss: 4.9871e-04 - val_loss: 7.4646e-04
Epoch 34/100
57/57 [==============================] - 2s 28ms/step - loss: 4.9541e-04 - val_loss: 8.9639e-04
Epoch 35/100
57/57 [==============================] - 2s 28ms/step - loss: 4.5409e-04 - val_loss: 0.0015
Epoch 36/100
57/57 [==============================] - 2s 28ms/step - loss: 4.2745e-04 - val_loss: 0.0010
Epoch 37/100
57/57 [==============================] - 2s 28ms/step - loss: 4.5775e-04 - val_loss: 0.0019
Epoch 38/100
57/57 [==============================] - 2s 29ms/step - loss: 4.6556e-04 - val_loss: 5.7972e-04
Epoch 39/100
57/57 [==============================] - 2s 28ms/step - loss: 4.4506e-04 - val_loss: 6.3251e-04
Epoch 40/100
57/57 [==============================] - 2s 28ms/step - loss: 4.3470e-04 - val_loss: 4.8006e-04
Epoch 41/100
57/57 [==============================] - 2s 28ms/step - loss: 3.9713e-04 - val_loss: 9.8011e-04
Epoch 42/100
57/57 [==============================] - 2s 29ms/step - loss: 4.3144e-04 - val_loss: 0.0030
Epoch 43/100
57/57 [==============================] - 2s 29ms/step - loss: 4.4934e-04 - val_loss: 5.1773e-04
Epoch 44/100
57/57 [==============================] - 2s 29ms/step - loss: 4.3007e-04 - val_loss: 5.0296e-04
Epoch 45/100
57/57 [==============================] - 2s 30ms/step - loss: 3.9985e-04 - val_loss: 0.0016
Epoch 46/100
57/57 [==============================] - 2s 31ms/step - loss: 4.2366e-04 - val_loss: 0.0011
Epoch 47/100
57/57 [==============================] - 2s 29ms/step - loss: 3.9213e-04 - val_loss: 0.0011
Epoch 48/100
57/57 [==============================] - 2s 26ms/step - loss: 4.0857e-04 - val_loss: 6.3791e-04
Epoch 49/100
57/57 [==============================] - 1s 25ms/step - loss: 3.8211e-04 - val_loss: 6.1667e-04
Epoch 50/100
57/57 [==============================] - 1s 25ms/step - loss: 3.7417e-04 - val_loss: 6.2659e-04
9/9 [==============================] - 1s 8ms/step
Test MSE: 374.8673
Test MAE: 15.0888
Test MAPE: 0.84%
Test R²: 0.9548

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import plotly.express as px
from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_absolute_percentage_error, mean_squared_error
import tensorflow as tf
from keras import Model
from keras.layers import Input, Dense, Dropout, LSTM
from keras.callbacks import EarlyStopping
import warnings

warnings.filterwarnings('ignore')

# === 数据加载与预处理 ===
df = pd.read_csv('Gold Price (2013-2022).csv')
df.drop(['Vol.', 'Change %'], axis=1, inplace=True)

# 转换日期格式并排序
df['Date'] = pd.to_datetime(df['Date'])
df.sort_values(by='Date', ascending=True, inplace=True)
df.reset_index(drop=True, inplace=True)

# 去除“,”符号并转换为float
NumCols = df.columns.drop(['Date'])
df[NumCols] = df[NumCols].replace({',': ''}, regex=True).astype('float64')

# 缺失值检测
if df.isnull().sum().sum() > 0:
    print("数据中存在缺失值！请检查。")
else:
    print("数据无缺失值。")

# 可视化黄金价格
fig = px.line(y=df.Price, x=df.Date, title="Gold Price History Data")
fig.update_traces(line_color='black')
fig.update_layout(plot_bgcolor='rgba(255,223,0,0.8)', xaxis_title="Date", yaxis_title="Scaled Price")
fig.show()

# === 数据切分 ===
test_size = df[df.Date.dt.year == 2022].shape[0]
scaler = MinMaxScaler()
scaler.fit(df.Price.values.reshape(-1, 1))

train_data = scaler.transform(df.Price[:-test_size].values.reshape(-1, 1))
test_data = scaler.transform(df.Price[-test_size - 60:].values.reshape(-1, 1))

# 滑动窗口构造
window_size = 60
X_train, y_train, X_test, y_test = [], [], [], []

for i in range(window_size, len(train_data)):
    X_train.append(train_data[i - window_size:i, 0])
    y_train.append(train_data[i, 0])

for i in range(window_size, len(test_data)):
    X_test.append(test_data[i - window_size:i, 0])
    y_test.append(test_data[i, 0])

X_train, y_train = np.array(X_train), np.array(y_train)
X_test, y_test = np.array(X_test), np.array(y_test)

# 调整形状
X_train = X_train.reshape((X_train.shape[0], X_train.shape[1], 1))
X_test = X_test.reshape((X_test.shape[0], X_test.shape[1], 1))
y_train = y_train.reshape((-1, 1))
y_test = y_test.reshape((-1, 1))

# === 模型构建 ===
def define_model():
    input1 = Input(shape=(window_size, 1))
    x = LSTM(units=64, return_sequences=True)(input1)
    x = Dropout(0.2)(x)
    x = LSTM(units=32)(x)  # 减少过拟合风险
    x = Dropout(0.2)(x)
    dnn_output = Dense(1)(x)  # 改用线性激活函数
    
    model = Model(inputs=input1, outputs=[dnn_output])
    model.compile(loss='mean_squared_error', optimizer='Adam')  # 更适合回归的优化器
    return model

model = define_model()

# 训练
early_stopping = EarlyStopping(monitor='val_loss', patience=10, restore_best_weights=True)
history = model.fit(X_train, y_train, epochs=100, batch_size=32, validation_split=0.2, 
                    callbacks=[early_stopping], verbose=1)


# === 可视化结果 ===
y_test_true = scaler.inverse_transform(y_test)
y_test_pred = scaler.inverse_transform(y_pred)

plt.figure(figsize=(15, 6))
plt.plot(df['Date'].iloc[:-test_size], scaler.inverse_transform(train_data), color='black', lw=2)
plt.plot(df['Date'].iloc[-test_size:], y_test_true, color='blue', lw=2)
plt.plot(df['Date'].iloc[-test_size:], y_test_pred, color='red', lw=2)
plt.title('Model Performance on Gold Price Prediction', fontsize=15)
plt.xlabel('Date', fontsize=12)
plt.ylabel('Price', fontsize=12)
plt.legend(['Training Data', 'Actual Test Data', 'Predicted Test Data'], loc='upper left', prop={'size': 15})
plt.grid(color='white')
plt.show()

# === 模型评估 ===
y_pred = model.predict(X_test)

MAPE = mean_absolute_percentage_error(y_test, y_pred)
MSE = mean_squared_error(y_test, y_pred)
RMSE = np.sqrt(MSE)
print(f"MAPE: {MAPE * 100:.2f}%")
print(f"RMSE: {RMSE:.2f}")
MAE = np.mean(np.abs(y_test - y_pred))
print(f"MAE: {MAE:.2f}")
from sklearn.metrics import r2_score
R2 = r2_score(y_test, y_pred)
print(f"R² Score: {R2:.2f}")
SMAPE = np.mean(2 * np.abs(y_test - y_pred) / (np.abs(y_test) + np.abs(y_pred))) * 100
print(f"SMAPE: {SMAPE:.2f}%")
from sklearn.metrics import explained_variance_score
EVS = explained_variance_score(y_test, y_pred)
print(f"Explained Variance Score: {EVS:.2f}")
residuals = y_test - y_pred
plt.figure(figsize=(10, 5))
plt.hist(residuals, bins=50, color='purple', alpha=0.7)
plt.title('Residual Distribution')
plt.xlabel('Residuals')
plt.ylabel('Frequency')
plt.show()

数据无缺失值。

Epoch 1/100
57/57 [==============================] - 4s 32ms/step - loss: 0.0074 - val_loss: 0.0026
Epoch 2/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0017 - val_loss: 0.0030
Epoch 3/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0016 - val_loss: 0.0019
Epoch 4/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0016 - val_loss: 0.0014
Epoch 5/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0014 - val_loss: 0.0019
Epoch 6/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0015 - val_loss: 0.0097
Epoch 7/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0013 - val_loss: 0.0017
Epoch 8/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0012 - val_loss: 0.0040
Epoch 9/100
57/57 [==============================] - 1s 23ms/step - loss: 0.0011 - val_loss: 0.0014
Epoch 10/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0011 - val_loss: 0.0027
Epoch 11/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0011 - val_loss: 0.0013
Epoch 12/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0011 - val_loss: 0.0012
Epoch 13/100
57/57 [==============================] - 1s 24ms/step - loss: 9.5916e-04 - val_loss: 0.0029
Epoch 14/100
57/57 [==============================] - 1s 24ms/step - loss: 9.7136e-04 - val_loss: 0.0019
Epoch 15/100
57/57 [==============================] - 1s 24ms/step - loss: 0.0010 - val_loss: 0.0026
Epoch 16/100
57/57 [==============================] - 1s 24ms/step - loss: 8.5942e-04 - val_loss: 0.0035
Epoch 17/100
57/57 [==============================] - 1s 24ms/step - loss: 8.0067e-04 - val_loss: 0.0011
Epoch 18/100
57/57 [==============================] - 1s 24ms/step - loss: 7.9242e-04 - val_loss: 7.1610e-04
Epoch 19/100
57/57 [==============================] - 1s 24ms/step - loss: 7.2196e-04 - val_loss: 8.6864e-04
Epoch 20/100
57/57 [==============================] - 1s 24ms/step - loss: 7.6009e-04 - val_loss: 6.9524e-04
Epoch 21/100
57/57 [==============================] - 1s 24ms/step - loss: 7.4915e-04 - val_loss: 0.0013
Epoch 22/100
57/57 [==============================] - 1s 24ms/step - loss: 7.0749e-04 - val_loss: 0.0036
Epoch 23/100
57/57 [==============================] - 1s 23ms/step - loss: 6.6179e-04 - val_loss: 0.0012
Epoch 24/100
57/57 [==============================] - 1s 24ms/step - loss: 6.3654e-04 - val_loss: 7.1714e-04
Epoch 25/100
57/57 [==============================] - 1s 24ms/step - loss: 5.8780e-04 - val_loss: 6.5300e-04
Epoch 26/100
57/57 [==============================] - 1s 24ms/step - loss: 5.6962e-04 - val_loss: 8.3446e-04
Epoch 27/100
57/57 [==============================] - 1s 24ms/step - loss: 6.0373e-04 - val_loss: 7.6859e-04
Epoch 28/100
57/57 [==============================] - 1s 25ms/step - loss: 5.4184e-04 - val_loss: 5.7977e-04
Epoch 29/100
57/57 [==============================] - 1s 24ms/step - loss: 5.6801e-04 - val_loss: 0.0021
Epoch 30/100
57/57 [==============================] - 1s 24ms/step - loss: 5.7730e-04 - val_loss: 0.0035
Epoch 31/100
57/57 [==============================] - 1s 24ms/step - loss: 4.8358e-04 - val_loss: 0.0011
Epoch 32/100
57/57 [==============================] - 2s 27ms/step - loss: 4.8153e-04 - val_loss: 7.5185e-04
Epoch 33/100
57/57 [==============================] - 2s 27ms/step - loss: 4.8292e-04 - val_loss: 0.0011
Epoch 34/100
57/57 [==============================] - 2s 27ms/step - loss: 4.9850e-04 - val_loss: 8.0108e-04
Epoch 35/100
57/57 [==============================] - 2s 28ms/step - loss: 4.3630e-04 - val_loss: 4.8953e-04
Epoch 36/100
57/57 [==============================] - 2s 27ms/step - loss: 4.5397e-04 - val_loss: 7.6943e-04
Epoch 37/100
57/57 [==============================] - 2s 27ms/step - loss: 4.4764e-04 - val_loss: 0.0011
Epoch 38/100
57/57 [==============================] - 2s 28ms/step - loss: 4.2593e-04 - val_loss: 4.8475e-04
Epoch 39/100
57/57 [==============================] - 2s 28ms/step - loss: 4.4397e-04 - val_loss: 7.5298e-04
Epoch 40/100
57/57 [==============================] - 2s 28ms/step - loss: 3.9450e-04 - val_loss: 0.0016
Epoch 41/100
57/57 [==============================] - 2s 28ms/step - loss: 4.5061e-04 - val_loss: 0.0013
Epoch 42/100
57/57 [==============================] - 2s 27ms/step - loss: 4.4132e-04 - val_loss: 5.6962e-04
Epoch 43/100
57/57 [==============================] - 2s 28ms/step - loss: 4.3400e-04 - val_loss: 8.6920e-04
Epoch 44/100
57/57 [==============================] - 2s 27ms/step - loss: 3.8915e-04 - val_loss: 0.0015
Epoch 45/100
57/57 [==============================] - 2s 27ms/step - loss: 3.9985e-04 - val_loss: 0.0016
Epoch 46/100
57/57 [==============================] - 2s 27ms/step - loss: 3.9131e-04 - val_loss: 4.3715e-04
Epoch 47/100
57/57 [==============================] - 2s 27ms/step - loss: 3.8516e-04 - val_loss: 6.2513e-04
Epoch 48/100
57/57 [==============================] - 2s 28ms/step - loss: 3.8856e-04 - val_loss: 6.3549e-04
Epoch 49/100
57/57 [==============================] - 2s 27ms/step - loss: 4.0803e-04 - val_loss: 6.9909e-04
Epoch 50/100
57/57 [==============================] - 2s 27ms/step - loss: 4.0064e-04 - val_loss: 0.0022
Epoch 51/100
57/57 [==============================] - 2s 27ms/step - loss: 3.6760e-04 - val_loss: 4.0346e-04
Epoch 52/100
57/57 [==============================] - 2s 27ms/step - loss: 3.7670e-04 - val_loss: 8.0508e-04
Epoch 53/100
57/57 [==============================] - 2s 27ms/step - loss: 3.9973e-04 - val_loss: 0.0010
Epoch 54/100
57/57 [==============================] - 2s 28ms/step - loss: 3.3143e-04 - val_loss: 3.8222e-04
Epoch 55/100
57/57 [==============================] - 2s 28ms/step - loss: 3.7577e-04 - val_loss: 4.4995e-04
Epoch 56/100
57/57 [==============================] - 2s 27ms/step - loss: 3.6030e-04 - val_loss: 0.0017
Epoch 57/100
57/57 [==============================] - 2s 28ms/step - loss: 3.3792e-04 - val_loss: 0.0012
Epoch 58/100
57/57 [==============================] - 1s 24ms/step - loss: 3.3102e-04 - val_loss: 0.0010
Epoch 59/100
57/57 [==============================] - 1s 24ms/step - loss: 3.4371e-04 - val_loss: 6.0616e-04
Epoch 60/100
57/57 [==============================] - 1s 24ms/step - loss: 3.9870e-04 - val_loss: 0.0028
Epoch 61/100
57/57 [==============================] - 1s 24ms/step - loss: 3.2062e-04 - val_loss: 8.2139e-04
Epoch 62/100
57/57 [==============================] - 1s 24ms/step - loss: 3.1101e-04 - val_loss: 0.0014
Epoch 63/100
57/57 [==============================] - 1s 24ms/step - loss: 3.7520e-04 - val_loss: 6.1046e-04
Epoch 64/100
57/57 [==============================] - 1s 24ms/step - loss: 3.2823e-04 - val_loss: 6.9230e-04

9/9 [==============================] - 0s 8ms/step
MAPE: 1.88%
RMSE: 0.02
MAE: 0.01
R² Score: 0.96
SMAPE: 1.87%
Explained Variance Score: 0.96

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import plotly.express as px
from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_absolute_percentage_error, mean_squared_error
from keras.models import Sequential
from keras.layers import LSTM, Dense, Dropout
from keras.callbacks import EarlyStopping
import warnings

warnings.filterwarnings('ignore')

# === 数据加载与预处理 ===
# 读取新数据
df = pd.read_csv('XAU_1d_data_2004_to_2024-09-20.csv')  # 替换为你的文件路径
df.columns = ['Date', 'Time', 'Open', 'High', 'Low', 'Close', 'Volume']  # 确保列名正确

# 合并日期和时间列，并转换为datetime格式
df['Datetime'] = pd.to_datetime(df['Date'] + ' ' + df['Time'])
df.sort_values(by='Datetime', ascending=True, inplace=True)
df.reset_index(drop=True, inplace=True)

# 转换为浮点数
NumCols = ['Open', 'High', 'Low', 'Close', 'Volume']
df[NumCols] = df[NumCols].replace({',': ''}, regex=True).astype('float64')

# 缺失值检测
if df.isnull().sum().sum() > 0:
    print("数据中存在缺失值！请检查。")
else:
    print("数据无缺失值。")

# 可视化黄金价格走势
fig = px.line(y=df['Close'], x=df['Datetime'], title="Gold Price Historical Data")
fig.update_traces(line_color='blue')
fig.update_layout(plot_bgcolor='rgba(255,255,255,0.8)', xaxis_title="Datetime", yaxis_title="Price")
fig.show()

# === 数据切分 ===
# 设置训练集和测试集划分时间点
test_size = int(0.2 * len(df))  # 测试集为数据的20%
scaler = MinMaxScaler()
scaler.fit(df['Close'].values.reshape(-1, 1))

train_data = scaler.transform(df['Close'][:-test_size].values.reshape(-1, 1))
test_data = scaler.transform(df['Close'][-test_size - 60:].values.reshape(-1, 1))

# 滑动窗口构造
window_size = 60
X_train, y_train, X_test, y_test = [], [], [], []

for i in range(window_size, len(train_data)):
    X_train.append(train_data[i - window_size:i, 0])
    y_train.append(train_data[i, 0])

for i in range(window_size, len(test_data)):
    X_test.append(test_data[i - window_size:i, 0])
    y_test.append(test_data[i, 0])

X_train, y_train = np.array(X_train), np.array(y_train)
X_test, y_test = np.array(X_test), np.array(y_test)

# 调整形状
X_train = X_train.reshape((X_train.shape[0], X_train.shape[1], 1))
X_test = X_test.reshape((X_test.shape[0], X_test.shape[1], 1))

# === 模型构建 ===
def build_model():
    model = Sequential()
    model.add(LSTM(units=64, return_sequences=True, input_shape=(X_train.shape[1], 1)))
    model.add(Dropout(0.2))
    model.add(LSTM(units=32, return_sequences=False))
    model.add(Dropout(0.2))
    model.add(Dense(1))
    model.compile(optimizer='adam', loss='mean_squared_error')
    return model

model = build_model()

# 训练
early_stopping = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)
history = model.fit(X_train, y_train, epochs=50, batch_size=32, validation_split=0.2, callbacks=[early_stopping], verbose=1)

# === 模型评估 ===
y_pred = model.predict(X_test)

MAPE = mean_absolute_percentage_error(y_test, y_pred)
MSE = mean_squared_error(y_test, y_pred)
RMSE = np.sqrt(MSE)
print(f"MAPE: {MAPE * 100:.2f}%")
print(f"RMSE: {RMSE:.2f}")

# === 可视化结果 ===
y_test_true = scaler.inverse_transform(y_test.reshape(-1, 1))
y_test_pred = scaler.inverse_transform(y_pred)

plt.figure(figsize=(15, 6))
plt.plot(df['Datetime'][:-test_size], scaler.inverse_transform(train_data), color='black', lw=2)
plt.plot(df['Datetime'][-test_size:], y_test_true, color='blue', lw=2)
plt.plot(df['Datetime'][-test_size:], y_test_pred, color='red', lw=2)
plt.title('Gold Price Prediction Model Performance', fontsize=15)
plt.xlabel('Datetime', fontsize=12)
plt.ylabel('Price', fontsize=12)
plt.legend(['Training Data', 'Actual Test Data', 'Predicted Test Data'], loc='upper left', prop={'size': 15})
plt.grid(color='white')
plt.show()

数据无缺失值。

Epoch 1/50
103/103 [==============================] - 5s 28ms/step - loss: 0.0085 - val_loss: 5.7880e-04
Epoch 2/50
103/103 [==============================] - 2s 24ms/step - loss: 0.0016 - val_loss: 4.0505e-04
Epoch 3/50
103/103 [==============================] - 2s 24ms/step - loss: 0.0013 - val_loss: 4.0723e-04
Epoch 4/50
103/103 [==============================] - 2s 24ms/step - loss: 0.0014 - val_loss: 5.3618e-04
Epoch 5/50
103/103 [==============================] - 2s 24ms/step - loss: 0.0012 - val_loss: 2.9975e-04
Epoch 6/50
103/103 [==============================] - 3s 24ms/step - loss: 0.0012 - val_loss: 2.1895e-04
Epoch 7/50
103/103 [==============================] - 3s 24ms/step - loss: 0.0011 - val_loss: 1.8931e-04
Epoch 8/50
103/103 [==============================] - 3s 24ms/step - loss: 9.8878e-04 - val_loss: 1.8997e-04
Epoch 9/50
103/103 [==============================] - 3s 25ms/step - loss: 9.9088e-04 - val_loss: 2.2553e-04
Epoch 10/50
103/103 [==============================] - 3s 25ms/step - loss: 9.3725e-04 - val_loss: 5.0751e-04
Epoch 11/50
103/103 [==============================] - 2s 24ms/step - loss: 8.9388e-04 - val_loss: 2.2041e-04
Epoch 12/50
103/103 [==============================] - 2s 24ms/step - loss: 8.4638e-04 - val_loss: 3.3183e-04
33/33 [==============================] - 1s 8ms/step
MAPE: 2.06%
RMSE: 0.02

from sklearn.metrics import r2_score, mean_absolute_error
# 模型预测
y_pred = model.predict(X_test)

# 反标准化
y_test_true = scaler.inverse_transform(y_test.reshape(-1, 1))
y_test_pred = scaler.inverse_transform(y_pred)

# 评估指标
MAPE = mean_absolute_percentage_error(y_test, y_pred)
MSE = mean_squared_error(y_test, y_pred)
RMSE = np.sqrt(MSE)
MAE = mean_absolute_error(y_test, y_pred)
R2 = r2_score(y_test, y_pred)

print(f"MAPE: {MAPE * 100:.2f}%")
print(f"MSE: {MSE:.2f}")
print(f"RMSE: {RMSE:.2f}")
print(f"MAE: {MAE:.2f}")
print(f"R^2 Score: {R2:.2f}")

33/33 [==============================] - 0s 8ms/step
MAPE: 2.06%
MSE: 0.00
RMSE: 0.02
MAE: 0.01
R^2 Score: 0.95