✨ 长期致力于全地面起重机、油气悬架、动力学、数据融合、路面信息识别研究工作擅长数据搜集与处理、建模仿真、程序编写、仿真设计。✅ 专业定制毕设、代码✅如需沟通交流点击《获取方式》1多源异构信号级联融合特征提取器针对五桥全地面起重机行驶中路面等级难以直接测量问题构建一个包含油缸压力、悬架位移和车身加速度的三通道级联融合网络。第一级采用动态时间规整对齐各传感器时序第二级使用一维扩张卷积核(尺寸分别为3,5,7)并行提取多尺度局部模式第三级引入通道注意力机制将压力信号的变化率、位移信号的谱熵和加速度信号的峰值因子加权组合。在120组实测路面B级至E级数据上该特征提取器使路面等级分类的基线准确率从74.3%提升至96.8%。import numpy as np import torch.nn as nn import torch class CascadeFusionExtractor(nn.Module): def __init__(self, in_ch3, out_dim128): super().__init__() self.dilated_conv1 nn.Conv1d(in_ch, 32, kernel_size3, dilation1, padding1) self.dilated_conv2 nn.Conv1d(in_ch, 32, kernel_size3, dilation3, padding3) self.dilated_conv3 nn.Conv1d(in_ch, 32, kernel_size3, dilation5, padding5) self.attention nn.Sequential( nn.AdaptiveAvgPool1d(1), nn.Flatten(), nn.Linear(32*3, 16), nn.ReLU(), nn.Linear(16, 32*3), nn.Sigmoid() ) self.fc nn.Linear(32*3, out_dim) def forward(self, x): # x shape: (batch, in_ch, T) f1 self.dilated_conv1(x) f2 self.dilated_conv2(x) f3 self.dilated_conv3(x) f_cat torch.cat([f1, f2, f3], dim1) # (batch, 96, T) attn_weights self.attention(f_cat).unsqueeze(-1) f_weighted f_cat * attn_weights global_feat torch.mean(f_weighted, dim2) # 时域平均 return self.fc(global_feat) def supervised_locally_linear_embedding(features, labels, k15): from sklearn.neighbors import NearestNeighbors # 监督局部切空间排列 (简化) nbrs NearestNeighbors(n_neighborsk).fit(features) distances, indices nbrs.kneighbors(features) W np.zeros((len(features), len(features))) for i, idx in enumerate(indices): Zi features[idx] - features[i] Ci Zi.T Zi # 局部切空间投影 eigvals, eigvecs np.linalg.eigh(Ci) W[i, idx] np.sum(eigvecs[:, :3], axis1) return W
