SAMSE 2018 IOP Conf. Series: Materials Science and Engineering 490 (2019) 062027
IOP Publishing doi:10.1088/1757-899X/490/6/062027
of papermaking wastewater, when the 6t h pivot element is extracted during calculation, namely, 6 d , the cumulative information contribution rate is 92.63%. Thus, the six pivot element are determined as the input vectors of the subsequent PSO-LSSVM model. As described in Section 2.2.2 above, the optimization process for the algorithm of parameters J and V PSO under the LSSVM method is shown in Figs. 3 and 4 after the RBF radial basis kernel function was selected, and the population particle number was set at 30, the minimum inertia weight was min w =0.01, the maximum inertia weight was min w =0.99, the particle maximum velocity was 2 max v , the particle minimum velocity was -2 min v , and the learning factor was 2 2 1 c c and the maximum number of iterations was 100. When J =0.3356 and V =2.2026, the RMSE of the objective function observation sample set reached the minimum, thereby it was determined as the optimal value of the parameter under the LSSVM method.
Fig 3. Regularization factor optimizing curve using PSO
Fig 4. Kernel parameter optimizing curve using PSO After the optimization for the input variable and parameter of the model was completed, the OCS-PCA-PSO-LSSVM model was applied to the test sample set to detect the model’s generalization ability. Table 1 shows the test results of different performance indicators for the three models of OCS-PCA-PSO-LSSVM, PCA-PSO-LSSVM and SVM. It may be observed from the table that the values of the maximum deviation, the maximum relative deviation, the average absolute deviation, the average relative deviation, the root mean square error, and the standard deviation of the OCS-PCA-PSO-LSSVM model are significantly lower than the corresponding results of the PCA-PSO-LSSVM model and the SVM model. Where, compared with the PCA-PSO-LSSVM model, the MAXE of the OCS-PCA-PSO-LSSVM model decreased by 39.15%, the MRE decreased by 25.00%, and the STD decreased by 29.89%. The reason for this is that when the PCA-PSO-LSSVM model predicted the 2 nd , 10 th , 20 th , and 21 st sample individuals in the test sample set, their prediction deviations were greater than their respective maximum fitting deviation the training sample set, so the OCS strategy was initiated 4 times to perform the iterative update of the model, therefore, from the 2 nd
7
Made with FlippingBook Digital Publishing Software