Table 2.

Test results of different activation function and number of convolution kernels (shaded group is the optimal structure; bold font indicates the optimal solution).

Group	Activation function	Number of convolution kernels	Training set		Validation set		Test set
			R2	RMSE	R2	RMSE	R2	RMSE
1	Relu	8, 16	0.9931	17.3	0.9860	31.4	0.9734	24.7
2		8, 32	0.9910	19.8	0.9888	22.1	0.9823	25.6
3		8, 64	0.9911	19.7	0.9892	21.7	0.9829	25.1
4		16, 32	0.9957	13.8	0.9876	23.2	0.9913	17.9
5		16, 64	0.9948	15.1	0.9908	20.0	0.9800	27.2
6		32, 64	0.9926	18.0	0.9867	24.1	0.9769	29.2
7	Tanh	8, 16	0.9314	54.7	0.9096	57.8	0.8710	75.0
8		8, 32	0.8904	69.2	0.8770	73.3	0.8795	66.8
9		8, 64	0.8892	53.4	0.8836	58.2	0.8854	58.0
10		16, 32	0.9439	49.5	0.8566	79.1	0.8625	71.3
11		16, 64	0.9377	40.9	0.9299	41.5	0.8947	56.8
12		32, 64	0.9248	57.3	0.8945	62.5	0.8891	69.5
13	Sigmoid	8, 16	0.9391	51.5	0.9229	53.4	0.9014	65.6
14		8, 32	0.9806	29.1	0.9749	33.1	0.9718	32.3
15		8, 64	0.9791	30.2	0.9644	39.4	0.9512	42.5
16		16, 32	0.9388	51.7	0.9214	53.9	0.8984	66.6
17		16, 64	0.9676	37.6	0.9630	40.2	0.9575	39.6
18		32, 64	0.9858	24.9	0.9714	35.3	0.9688	34.0