class MultiScaleSTBlock(nn.Module):
    def __init__(self, in_channels, out_channels, time_steps):
        super().__init__()
        self.gcn = GCNConv(in_channels, out_channels)
        self.stcnn = STConv2d(out_channels, out_channels, (3, time_steps))
        self.channel_attention = ChannelAttention(out_channels)

    def forward(self, x, edge_index):
        # 图卷积提取拓扑特征
        graph_feat = self.gcn(x, edge_index)

        # 时空卷积捕捉时空耦合
        batch_size, nodes, features = graph_feat.shape
        st_input = graph_feat.view(batch_size, features, nodes, 1)
        spatiotemporal_feat = self.stcnn(st_input)

        # 多尺度特征融合
        multi_scale_feat = torch.cat([graph_feat, spatiotemporal_feat], dim=1)
        return self.channel_attention(multi_scale_feat)

def risk_score(user_profile, spatiotemporal_embedding):
    # 用户特征: 年龄、基础疾病、出行记录
    user_vec = torch.cat([
        user_profile,
        torch.mean(spatiotemporal_embedding[user_trajectory], dim=0)
    ])

    # 风险评估
    risk = sigmoid(linear(user_vec))
    return risk > 0.7  # 高风险阈值