Parameter Selection of Support Vector Regression Based on Particle Swarm Optimization

Parameters selection of support vector machine is the key issue that impacts its accurate performance. A method for support vector regression machine with basic particle swarm optimization (BPSO) algorithm is proposed in this paper. Furthermore, in order to improve the efficiency of the PSO algorithm, a linear decreasing strategy is used to dynamically change the weight. So, an improve PSO (IPSO) algorithm was also proposed in this paper. Then, two different models using BPSO and IPSO respectively were used to forecast the density of the acid-lead battery electrolyte. The experimental results indicate that both BPSO and IPSO have high prediction accuracy and efficiency. the time of the parametric searching by IPSO is obviously decreased to that of BPSO. The mean squared error (MSE) of the prediction model using BPSO is about 2.46684×10^-4, Meanwhile, the MSE of the model using IPSO is only about 2.01948×10^-4. So, the IPSO algorithm has more superior performance on convergence speed and global optimization.