Page - 911 - in Book of Full Papers - Symposium Hydro Engineering

5 1 2 3 4 5 6 7 8 9 10 11 0.019 0.005 0.016 0.311 0.252 0.240 0 0. .2 6 95 0.109 0.050 0.06 39 8 Y X X X X X X X X X X X             [7] 3.3. RBF NEURAL NETWORK RBF neural network was created and the calling format is:   , ,net newrb PT spread [8] where P is the input item, T is the output item, and spread is the distribution density of radial basis functions. Five principal components are selected as the input items of the neural network:   1 2 3 4 5, , , , T Y Y Y Y YP [9] The displacements prediction data are chosen as the output item of the neural network, that is:   T pT y [10] 3.4. COMPARISON In order to verify the prediction effect, the PCA-RBF neural network model is compared with the multivariate regression analysis statistical model in this section. The above-mentioned 126 sets of data are used as the sample, and the multivariate regression function is shown as follow: 1 2 3 4 5 6 7 8 9 10 11 0.449 0.157 0.002 3.49 05 0.227 0.092 0.354 0.191 0.15 0.147 0.919 0.812 sY X X e X X X X X X X X X              [11] The training and prediction results of the two models are shown in Fig. 2. The determination coefficient (R2), mean-square error (RMSE) and mean absolute deviation (MAE) are defined as (12) ~ (14), where ( )sy i is prediction value and ( )y i is measured value. The results are listed in Table 3, and the comparison results of the residuals is shown in Fig. 2. 2 2 1 2 2 1 1 [ ( ( ) )( ( ) )] ( ( ) ) ( ( ) ) N s si N N s si i y i y y i y R y i y y i y            [12] 911