改成pytorch库

在PyTorch库中，我们可以使用类似的方式创建一个CNN-LSTM模型。以下是使用PyTorch的代码示例：

import torch
import torch.nn as nn
import torch.optim as optim

# 假设数据已经在 PyTorch 的 Tensor 形式下
wind_speed_tensor = ...  # 风速数据
radiation_tensor = ...  # 辐照度数据
temperature_tensor = ...  # 温度数据

class捆打包合Model(nn.Module):
    def __init__(self, input_channels):
        super().__init__()
        self.conv_layer = nn.Conv1d(input_channels, 32, kernel_size=3, padding=(3-1)//2)  # 注意填充以保持输出尺寸不变
        self.pooling = nn.MaxPool1d(2)
        self.lstm_layer_1 = nn.LSTM(32, 64, bidirectional=True)
        self.lstm_layer_2 = nn.LSTM(128, 32, bidirectional=True)  # 上一层是双向LSTM，所以输出维度翻倍
        self.fc = nn.Linear(32, 1)  # 输出层

    def forward(self, x):
        x = self.conv_layer(x)
        x = self.pooling(x)
        x, _ = self.lstm_layer_1(x)  # 只保留最后一个时间步的隐藏状态
        x = self.lstm_layer_2(x)[0]  # 又是一次取最后一个时间步
        x = x.mean(dim=1)  # 求平均值以获得全局特征
        x = self.fc(x)
        return x

# 初始化模型、损失函数和优化器
input_channels = wind_speed_data_dim + radiation_data_dim
model = 拢打包合Model(input_channels)
criterion = nn.MSELoss()  # 由于是回归任务，使用MSE Loss
optimizer = optim.Adam(model.parameters(), lr=0.001)

# 训练过程
for epoch in range(num_epochs):
    optimizer.zero_grad()
    output = model(torch.cat((wind_speed_tensor, radiation_tensor), dim=1))
    loss = criterion(output, temperature_tensor)
    loss.backward()
    optimizer.step()

# 测试和预测
predictions = model(torch.cat((wind_speed_tensor_test, radiation_tensor_test), dim=1))

#