New algorithm used hybrid coding, that is, taking the binary encoding method to encode the neural network structure and taking the real number encoding method to encode the weights between hidden Bicalutamide Cosudex layer and output layer, so that we can achieve the self-adaptation of adjusting the structure of neural network and the learning of connection weight simultaneously. A good structure has been got; however, the weight optimization is incomplete; it needs to be further optimized. Least mean square (LMS) algorithm [14–16] is chosen,

to optimize the connection weights continuously. Finally, a precise RBF neural network has been obtained. To verity the validity of the new algorithm, this study arranges two experiments, using three UCI standard data sets to test. From the following, some aspects to evaluate the algorithm, such as success training rate, training step, and recognition accuracy rate, are obtained. By comparing

with every experiment results, it verifies the superiority of the new optimizing algorithm. 2. Genetic Algorithm and RBF Neural Network 2.1. The Basic Theory of Genetic Algorithm Genetic algorithm starts from a population of represented potential solution set; however, the population is composed of a certain number of encoded gene individuals, which is the entities with characteristic chromosome. The main problems of constructing the genetic algorithm are the solvable encoding method and the design of genetic operator. Faced with different optimization methods, we need to use different encoding method and genetic operators of different operation, so they as well as the degree of the understanding of the problems to be solved are the main point determining whether the application of genetic algorithm can succeed. It is an iterative procedure; in each iteration, it retains a candidate solution and sorts them by the quality of the solutions and then chooses some of the solution according some indicators and uses genetic operators to compute it to produce a new generation of candidate solutions. We will repeat this process until it meets some convergence index Figure 1 clearly shows the process of the genetic algorithm.

Figure 1 The flow chart of genetic Brefeldin_A algorithm. 2.2. The Basic Theory of RBF Neural Network The work thought of RBF network is to take RBF as the “basis” of the hidden layer units, so as to construct the hidden layer space. It is a nonlinear function that is symmetrical on the central points and distributed locally, when the central points of the RBF are determined; then the input vector can be directly mapped to the hidden space. But the mapping from the hidden space to the output space is linear, that is, the linear weighting sum of the network unit output; the weight here is the network’s adjustable parameters. The RBF network is a three-layer feed-forward network which is composed of input layer, hidden layer, and output layer.