|
Learning is a basic human intelligent activities, learning ability is the fundamental characteristic of human intelligence. Machine learning machine simulation in the artificial intelligence systems and process to achieve a variety of learning behavior. The traditional machine learning methods include neural networks, wavelet networks, fuzzy systems, and built on a new statistical learning theory based machine learning methods - support vector machine. Radial basis function neural network is a novel and effective feedforward neural networks, it has other forward network does not have the best approximation of the performance and characteristics of the global optimum, and the structure is simple, fast training. RBF neural network, the number and location of the center of the hidden layer is the key to the merits of the overall network performance, the direct impact on the performance of the network. The number of centers i.e. the number of nodes of the hidden layer is selected to be much easily lead to over-fitting, so that the ability to promote decline; center number is selected to be too little, the learning network is insufficient learning of the information contained in the sample, will make the promotion reduced ability. In practical applications, the advantage of the RBF network linear learning algorithm to complete the previous work done by the non-linear learning algorithm, neural network while maintaining the characteristics of the nonlinear algorithm has high accuracy. However, in addressing the problem of high-dimensional data, the traditional way to determine the RBF network generalization ability has obvious shortcomings. Solid mathematical theory of support vector machine algorithm based on statistical learning theory and rigorous theoretical analysis, a complete theory, global optimization, adaptability, good generalization ability, etc., it largely solve the machine learning model selection and learning, non-linear, the curse of dimensionality, local minimum points, due to the support vector machine in pattern recognition, regression estimation, function approximation, risk budgeting, financial sequence analysis, density estimation, novelty test was a great success in various fields, it immediately became the object of study of machine learning, neural networks, artificial intelligence, such as the direction of experts and scholars. It uses structural risk minimization principle, comprehensive statistical learning, machine learning and neural networks technology, and at the same time minimize the empirical risk, improve the algorithm generalization capability. It is compared with traditional machine learning methods have good potential application and development prospects. In this paper, radial basis function neural network and support vector machine is the main object of study, the basis of the theory of machine learning methods as well as the mechanism of RBF neural networks and support vector machines, the analysis of the internal relations of these two learning methods. In this paper, on the basis of the processes and the basic principles of the study this interconnectedness and elaborated genetic algorithm, RBF neural network optimization algorithm, genetic algorithm parameter selection for support vector machine model, based on support vector machines and genetic algorithm reuse the establishment of a support vector machine to construct the RBF neural network. This algorithm avoids the traditional algorithm easy to fall into the shortcomings of the local minimum point, and does not require pre-specified network structure by a large number of experiments or empirically. Finally, use the RBF neural network algorithm to optimize for the identification of nonlinear systems through the simulation results show that the RBF network has better recognition accuracy and generalization ability.
|