python代码Bi-LSTM、CNN-BiLSTM

代号跳

已于 2023-12-27 12:43:58 修改

阅读量1.8k

点赞数 14

文章标签： python lstm cnn 人工智能

于 2023-12-27 12:40:40 首次发布

本文链接：https://blog.csdn.net/m0_54184704/article/details/135242546

版权

文章目录

一、Bi-LSTM

时间序列单变量预测

模型

import numpy as np
import tensorflow as tf
gpus = tf.config.list_physical_devices("GPU")

if gpus:
    tf.config.experimental.set_memory_growth(gpus[0], True)  #设置GPU显存用量按需使用
    tf.config.set_visible_devices([gpus[0]],"GPU")

from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error
from math import sqrt
import matplotlib.pyplot as plt
import pandas as pd

# 支持中文
plt.rcParams['font.sans-serif'] = ['SimHei']  # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False  # 用来正常显示负号

# 确保结果尽可能重现
from numpy.random          import seed
seed(1)
tf.random.set_seed(1)

#导入数据
oil=pd.read_csv('D:/April/postgraduate/研1/课程/时间序列分析/期末论文/china_oil.csv')
data=oil
oil


from tensorflow.keras.regularizers import l2

# 数据归一化
scaler = MinMaxScaler(feature_range=(-1, 1))
data_normalized = scaler.fit_transform(data.values.reshape(-1, 1)).reshape(-1)

# 创建时间窗口
def create_sequence_data(data, time_steps):
    sequences = []
    labels = []
    for i in range(len(data) - time_steps):
        seq = data[i:i+time_steps]
        label = data[i+time_steps:i+time_steps+1]
        sequences.append(seq)
        labels.append(label)
    return np.array(sequences), np.array(labels)

time_steps = 3  # 可以根据需要调整时间窗口大小
X, y = create_sequence_data(data_normalized, time_steps)

# 划分训练集和验证集
val_size = 100
X_train, X_val = X[:-val_size], X[-val_size:]
y_train, y_val = y[:-val_size], y[-val_size:]


# 构建Bi-LSTM模型
model = tf.keras.Sequential([
    tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(50, activation='relu', kernel_regularizer=l2(0.01)), input_shape=(X_train.shape[1], 1)),
    
    tf.keras.layers.Dense(1)
])

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

# 训练模型
history = model.fit(X_train, y_train,
                    epochs=20,
                    batch_size=32, 
                    validation_data=(X_val, y_val),
                    verbose=2)

# 进行预测
predicted = model.predict(X_val)

# 反归一化
predicted = scaler.inverse_transform(predicted)
y_val_original = scaler.inverse_transform(y_val)

测试集效果图

# 绘制预测结果
plt.plot(predicted, label='Predicted')
plt.plot(y_val_original, label='True')
plt.title('price - Predicted vs True')
plt.legend()
plt.show()

loss曲线

# 获取训练过程中的损失值
train_loss = history.history['loss']
val_loss = history.history['val_loss']

# 绘制损失曲线
plt.plot(train_loss, label='Training Loss')
plt.plot(val_loss, label='Validation Loss')
plt.title('price Training and Validation Loss')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend()
plt.show()

模型评估

from sklearn.metrics import mean_squared_error, mean_absolute_error
from sklearn.utils import check_array
from sklearn.metrics import r2_score

# 使用RMSE评估模型性能
rmse = sqrt(mean_squared_error(y_val_original, predicted))
print(f'MSE: {round(rmse,4)}')

# 计算MAPE
def mean_absolute_percentage_error(y_true, y_pred): 
    y_true, y_pred = np.array(y_true), np.array(y_pred)
    return np.mean(np.abs((y_true - y_pred) / y_true)) * 100

mape = mean_absolute_percentage_error(y_val_original, predicted)
print(f'MAPE: {round(mape,4)}')

# 计算R-squared (R2)
r2 = r2_score(y_val_original, predicted)
print(f'R2: {round(r2,4)}')

# 使用RMSE评估模型性能
rmse = sqrt(mean_squared_error(y_val_original, predicted))
print(f'验证集RMSE: {round(rmse,4)}')
from sklearn.metrics import mean_absolute_error

# 计算MAE
mae = mean_absolute_error(y_val_original, predicted)
print("Mean Absolute Error: ", mae)

二、CNN-BiLSTM

模型

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.preprocessing import MinMaxScaler
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv1D, MaxPooling1D, Bidirectional, LSTM, Dense, Flatten

# 假设有一个名为data的DataFrame，其中包含你的电力负荷时间序列数据

# 数据归一化

scaler = MinMaxScaler(feature_range=(-1, 1))
data_normalized = scaler.fit_transform(data.values.reshape(-1, 1)).reshape(-1)

# 创建时间窗口
def create_sequence_data(data, time_steps):
    sequences = []
    labels = []
    for i in range(len(data) - time_steps):
        seq = data[i:i+time_steps]
        label = data[i+time_steps]
        sequences.append(seq)
        labels.append(label)
    return np.array(sequences), np.array(labels)

time_steps =10  # 可以根据需要调整时间窗口大小
X, y = create_sequence_data(data_normalized, time_steps)

# 划分训练集和验证集
val_size = 100  # 取后200个数据
X_train, X_val = X[:-val_size], X[-val_size:]
y_train, y_val = y[:-val_size], y[-val_size:]

# 调整 X_train 和 X_val 的形状以匹配模型输入
X_train = X_train.reshape(X_train.shape[0], X_train.shape[1], 1)
X_val = X_val.reshape(X_val.shape[0], X_val.shape[1], 1)

# 构建CNN-BiLSTM模型
model = Sequential([
    Conv1D(filters=64, kernel_size=1, activation='relu', input_shape=(X_train.shape[1], 1)),
    MaxPooling1D(pool_size=2),
    Bidirectional(LSTM(128, activation='tanh')),
    Dense(1)
])

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

# 训练模型
history = model.fit(X_train, y_train,
                    epochs=50,
                    batch_size=32, 
                    validation_data=(X_val, y_val),
                    verbose=2)

# 进行预测
predicted = model.predict(X_val)

# 反归一化
predicted = scaler.inverse_transform(predicted)
y_val_original = scaler.inverse_transform(y_val)

loss

# 绘制损失曲线
plt.plot(history.history['loss'], label='Training Loss')
plt.plot(history.history['val_loss'], label='Validation Loss')
plt.title('Training and Validation Loss')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend()
plt.show()

测试集效果

# 绘制预测结果
plt.plot(y_val_original, label='True')
plt.plot(predicted, label='Predicted')
plt.title('Validation Set - True vs Predicted')
plt.xlabel('Time')
plt.ylabel('Power Load')
plt.legend()
plt.show()

模型评估

# 计算模型评估指标
rmse = np.sqrt(mean_squared_error(y_val_original, predicted))
mape = np.mean(np.abs((y_val_original - predicted) / y_val_original)) 
r2 = r2_score(y_val_original, predicted)

# 打印模型评估指标
print(f'Root Mean Squared Error (RMSE): {rmse:.4f}')
print(f'Mean Absolute Percentage Error (MAPE): {mape:.4f}')
print(f'R-squared (R2): {r2:.2f}')
from sklearn.metrics import mean_absolute_error

# 计算MAE
mae = mean_absolute_error(y_val_original, predicted)
print("Mean Absolute Error: ", mae)

三、copula相关系数

import numpy as np
from scipy.stats import kendalltau
from sklearn.neighbors import KernelDensity

##函数 
def calculate_kendall_tau(X, Y, bandwidth=1.0, kernel='gaussian', sample_size=len(data), random_seed=45):
    """
    计算两个变量之间的Kendall Tau相关系数。

    参数:
    - X: 第一个变量的数据
    - Y: 第二个变量的数据
    - bandwidth: 核密度估计的带宽
    - kernel: 核函数的类型
    - sample_size: 从估计的分布中生成样本的大小，默认为输入数据的大小
    - random_seed: 随机种子，用于生成样本

    返回:
    - kendall_tau: Kendall Tau相关系数
    """

    # 将变量堆叠成一个2D数组
    data = np.column_stack((X, Y))

    # 拟合核密度估计模型
    kde = KernelDensity(bandwidth=bandwidth, kernel=kernel)
    kde.fit(data)

    # 从估计的分布中生成样本
    if sample_size is None:
        sample_size = len(data)

    if random_seed is not None:
        np.random.seed(random_seed)

    copula_samples = kde.sample(sample_size)

    # 计算Kendall Tau相关系数
    kendall_tau, _ = kendalltau(copula_samples[:, 0], copula_samples[:, 1])

    return kendall_tau

举例

no2=data['no2']
pm2_5=data['pm2_5']
pm10=data['pm10']
co=data['co']
o3=data['o3']
so2=data['so2']

# 示例用法
# 注意：请替换下面的 X_data 和 Y_data 为你的实际数据

result = calculate_kendall_tau(no2, Y)
print(f'no2 Kendall Tau 相关系数: {result:.2f}')