为时间序列预测的PyTorch LSTM模型添加CNN层遇维度错误求助

输入张量尺寸为[16, 15, 2]，其中16是批量大小，15是时间步，2是特征数，模型需支持1或3步未来预测。添加CNN层后出现报错：

RuntimeError: Given groups=1, weight of size [1, 2, 3], expected input[16, 15, 2] to have 2 channels, but got 15 channels instead

原模型代码如下：

class TorchCNN_LSTM(nn.Module):
    def __init__(self, input_dim, hidden_dim, h2 ,layer_dim, output_dim, dropout_prob):
        super(TorchCNN_LSTM, self).__init__()
        .....
        # CNN layer
        self.conv = nn.Conv1d(input_dim, output_dim,
                              kernel_size=(3,), padding=2)
        # LSTM layers
        self.lstm = nn.LSTM(
            input_dim, hidden_dim, layer_dim, batch_first=True, dropout=dropout_prob
        )
        # Fully connected layer
        self.fc_1 = nn.Linear(hidden_dim, h2)  # fully connected
        self.fc_2 = nn.Linear(h2, output_dim)  # fully connected last layer
        self.relu = nn.ReLU()

    def forward(self, x):

        #this is my trying to reshap. but id doesn't work.
        #x_out= x.view(len(x), 1, -1)
        #and
        #x = x.permute(0,2,1)

        # Initializing hidden state for first input with zeros
        h0 = torch.zeros(self.layer_dim, x.size(0), self.hidden_dim , device = x.device ).requires_grad_() 
        # Initializing cell state for first input with zeros
        c0 = torch.zeros(self.layer_dim, x.size(0), self.hidden_dim, 
             device=x.device).requires_grad_()

        out = self.conv(x_out)
        out = self.lstm(out, (h0.detach(), c0.detach()))
        out = out[:, -1, :]
        out = self.fc_1(out)  # first dense
        out = self.relu(out)  # relu
        out = self.fc_2(out)  # final output
        return out

错误核心原因 #

PyTorch的nn.Conv1d要求输入维度为[batch_size, channels, sequence_length]，但你的输入是[batch_size, sequence_length, channels]，导致Conv1d误将时间步15识别为通道数，而实际通道数应为2，因此触发维度不匹配报错。

修正步骤 #

1. 适配Conv1d输入维度：用permute(0,2,1)将输入从[16,15,2]转为[16,2,15]，让Conv1d正确识别通道数。
2. 适配LSTM输入维度：CNN输出维度为[batch_size, output_channels, new_seq_len]，需再次用permute(0,2,1)转回[batch_size, new_seq_len, output_channels]，符合LSTMbatch_first=True的输入格式。
3. 修正LSTM输入参数：LSTM的输入维度应改为CNN的输出通道数（即output_dim），而非原输入特征数input_dim。

修正后的完整代码 #

import torch
import torch.nn as nn

class TorchCNN_LSTM(nn.Module):
    def __init__(self, input_dim, hidden_dim, h2, layer_dim, output_dim, dropout_prob):
        super(TorchCNN_LSTM, self).__init__()
        self.layer_dim = layer_dim
        self.hidden_dim = hidden_dim
        
        # CNN层：input_dim为输入通道数，输出output_dim个通道
        self.conv = nn.Conv1d(input_dim, output_dim, kernel_size=3, padding=2)
        # LSTM输入维度改为CNN的输出通道数
        self.lstm = nn.LSTM(
            output_dim, hidden_dim, layer_dim, batch_first=True, dropout=dropout_prob
        )
        # 全连接层
        self.fc_1 = nn.Linear(hidden_dim, h2)
        self.fc_2 = nn.Linear(h2, output_dim)
        self.relu = nn.ReLU()

    def forward(self, x):
        # 转换维度适配Conv1d：[batch, seq_len, channels] -> [batch, channels, seq_len]
        x_conv = x.permute(0, 2, 1)
        # CNN前向传播，输出维度：[batch, output_dim, new_seq_len]
        out = self.conv(x_conv)
        # 转换维度适配LSTM：[batch, output_dim, new_seq_len] -> [batch, new_seq_len, output_dim]
        out = out.permute(0, 2, 1)
        
        # 初始化隐藏状态与细胞状态
        h0 = torch.zeros(self.layer_dim, x.size(0), self.hidden_dim, device=x.device)
        c0 = torch.zeros(self.layer_dim, x.size(0), self.hidden_dim, device=x.device)
        
        # LSTM前向传播，仅保留输出张量
        out, _ = self.lstm(out, (h0.detach(), c0.detach()))
        # 取最后一个时间步的输出用于预测
        out = out[:, -1, :]
        
        # 全连接层输出
        out = self.fc_1(out)
        out = self.relu(out)
        out = self.fc_2(out)
        return out

维度流转验证 #

• 输入x：[16,15,2]
• 适配Conv1d后：[16,2,15]
• CNN输出：[16, output_dim, 17]（序列长度计算公式：原长度 + 2*padding - kernel_size +1）
• 适配LSTM后：[16,17, output_dim]
• LSTM最后时间步输出：[16, hidden_dim]
• 最终输出：[16, output_dim]（若需3步预测，将output_dim设为3即可）

内容的提问来源于stack exchange，提问作者Sadeq Al-Ahdal