【CEEMDAN-VMD-Transformer-LSTM】双重分解+Transformer-LSTM多变量时序预测

import numpy as np
import ceemdan
import vmd
from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import LSTM, Dense
from transformers import TransformerModel, TFAutoModel

生成示例数据

假设我们有一个包含多个变量的时间序列数据，存储在一个二维数组中

data = np.random.rand(100, 3) # 这里假设有3个变量，时间步长为100

数据预处理

scaler = MinMaxScaler()
scaled_data = scaler.fit_transform(data)

使用CEEMDAN进行信号分解

emd = ceemdan.CEEMDAN()
imfs = emd.ceemdan(scaled_data)

使用VMD进行信号分解

vmd = vmd.VMD()
modes = vmd.decompose(scaled_data.T)

将分解后的信号重新组合为训练集和测试集

这里假设我们使用最后一个分解模式作为目标变量

X = np.concatenate(imfs[:-1], axis=1)
y = modes[-1].T

划分训练集和测试集

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, shuffle=False)

使用Transformer进行特征提取

transformer_model = TFAutoModel.from_pretrained(“bert-base-uncased”)
transformer = TransformerModel(transformer_model)

X_train_transformed = transformer.predict(X_train)
X_test_transformed = transformer.predict(X_test)

定义LSTM模型

model = Sequential()
model.add(LSTM(64, input_shape=(X_train_transformed.shape[1], X_train_transformed.shape[2])))
model.add(Dense(y_train.shape[1]))
model.compile(loss=‘mean_squared_error’, optimizer=‘adam’)

训练模型

model.fit(X_train_transformed, y_train, epochs=100, batch_size=32, verbose=2)

在测试集上进行预测

predictions = model.predict(X_test_transformed)

将预测结果反向缩放

scaled_predictions = scaler.inverse_transform(predictions)

打印预测结果

print(scaled_predictions)