A Hybrid Algorithm Of PSO And ABC Used To Optimize The Parameters Of SVM And Its Application

Because of the limitations of particle swarm optimization (PSO) itself, the characteristics of other algorithms are needed to compensate. Therefore, the hybrid algorithm of PSO and other swarm intelligence optimization becomes one of the focus algorithms to study. Artificial bee colony (ABC) which is a new swarm intelligence algorithm, has the characteristics of both global and local optimum. With the combination of PSO, it can make up for the shortages of PSO which is easily falling into local optimum and premature convergence. Meanwhile, it can maintain the characteristics of global search and easy to implement of PSO. So this paper proposes an improved particle swarm algorithm and is used to optimize the SVM parameters which are applied to speech recognition.Firstly, this paper puts forward a parallel hybrid algorithm of PSO and ABC by combining artificial bee colony algorithm and particle swarm algorithm. The algorithm is applied to the function optimization problems with optimizing for minimum value of typical single-peak and multi-peak functions to test the optimization performance. The experimental results show that compared with the adaptive particle swarm optimization, the hybrid optimization algorithm has the characteristics of high optimization precision, high convergence accuracy, fewer iteration times and high success rate, which reflects good robustness and fast convergence speed. On the basis of the algorithm, this paper introduces the artificial bee colony algorithm with adjustment factors. The experimental results show that the hybrid algorithm with adjustment factors is more suitable for multi-apeak function in the high dimension, which shows good optimization ability and avoids local optimum.Secondly, because the support vector machine, namely SVM, which is good at solving classification problem, is a new type of machine learning theory. The kernel parameters of SVM have an important influence on the classification performance and the optimization method is still relatively limited. Therefore, this hybrid algorithm is used to optimize the kernel parameter of SVM and applied to speech recognition. The experimental results show that compared with the support vector machine with the kernel parameters optimized by PSO, the kernel parameters of support vector machine optimized using the hybrid algorithm has a good noise immunity, robustness and a strong speech recognition and generalization ability.Finally, because of the type of kernel function determines the classification performance of SVM. As a new kernel function, the mixed kernel which has the performance of both local kernel function and global kernel function. So improve the classification performance of the mixed kernel SVM by optimizing the parameters to get the optimal combination of parameters is one of the current research focuses. This paper uses the hybrid algorithm to optimize the mixed kernel parameters of SVM and the optimized SVM is applied to speech recognition. The experimental results show that compared with the mixed kernel of SVM optimized by PSO, the mixed kernel of SVM optimized by the hybrid algorithm has a high recognition rate and it can be able to adapt to the characteristics of different samples. Through the comprehensive comparison of the SVM classification performance by optimizing the mixed kernel and RBF kernel using the hybrid algorithm. The results show that the mixed kernel of SVM has good robustness and generalization ability.