Table 3.

Model network structural parameters

Index	Layer	Weight parameters	Trainable parameters
1	Conv2d (4, 8, kernel_size=(3, 1), stride=(2, 1))	4 ∗ 8 ∗ 3 ∗ 1 = 96	96
2	Conv2d (8, 16, kernel_size=(3, 1), stride=(2, 1))	8 ∗ 16 ∗ 3 ∗ 1 = 384	384
3	Conv2d (16, 64, kernel_size=(3, 1), stride=(2, 1))	16 ∗ 64 ∗ 3 ∗ 1 = 3,072	3,072
4	Linear (in_features = 64, out_features = 64)	64 ∗ 64 = 4,096	4,096
5	MultiheadAttention (out_proj = 64, 64)	64 ∗ 64 = 4,096	4,096
6	Linear (in_features = 64, out_features = 64)	64 ∗ 64 = 4,096	4,096
7	MultiheadAttention (out_proj = 64, 64)	64 ∗ 64 = 4,096	4,096
8	Linear (dense_soh: in_features = 64, out_features = 1)	64 ∗ 1 = 64	64