Table 1.

Overall performance comparison. SAE–DNN: deep neural network with stacked autoencoders; LR: liner regression; RFR: random forest regression; SVR: support vector regression; DNNs: deep neural networks; CNNs: convolutional neural networks.

Method. (Stage1 + Stage2)	MSE	${\mathit{R}}^2$	Method (Stage1 + Stage2)	MSE	${\mathit{R}}^2$
LR + CNNs	77.59	0.59	SVR + ConvGRU	56.14	0.72
LR + ConvLSTM	72.39	0.61	SVR + ResNet	49.58	0.77
LR + ConvGRU	70.65	0.63	DNNs + CNNs	59.54	0.72
LR + ResNet	65.23	0.65	DNNs + ConvLSTM	54.28	0.78
RFR + CNNs	59.75	0.62	DNNs + ConvGRU	54.69	0.79
RFR + ConvLSTM	55.96	0.66	DNNs + ResNet	45.98	0.81
RFR + ConvGRU	54.29	0.68	SAE-DNNs + CNNs	50.87	0.79
RFR + ResNet	47.68	0.72	SAE-DNNs + ConvLSTM	48.85	0.83
SVR + CNNs	65.49	0.68	SAE-DNNs + ConvGRU	49.58	0.83
SVR + ConvLSTM	55.97	0.7	SAE-DNNs + ResNet	40.25	0.86

The bold denotes to the best performance.