Intelligent Learning Algorithm Of RBF Neural Network And Its Application In Soft Sensor

The RBF neural network is used to build soft measurement model. Many parameters need to be addressed. In order to make model has good forecasting performance, in our actual modeling, we often introduce some intelligent optimization method to choose the appropriate parameter to participate in the modeling.The main contents are as follows:1. When using RBF neural network to model, of which the number of hidden layer and the related parameters will have an effect on the performance of the model. So we choose genetic algorithm and particle swarm algorithm to optimize the RBF neural network design. First of all, we need to take advantage of genetic algorithm and particle swarm algorithm to determine the model structure, then we have the good future, this future is our best neural network, the first to determine the structure and then the parameters of method can be regarded as a serial optimization.2. The speed of the model established by Serial optimization method is slow, because this method need genetic algorithm or particle swarm algorithm to generate two populations, one is used to determine the number of input variables and network among the number of hidden layer, and another is used to implement the optimization of network parameters, and optimization is carried out separately. In order to improve the operation efficiency of the models, we propose a parallel optimization method. The method of determination of model structure and network parameters is produced by a population, and the structural parameters of model and the optimization of network parameters was conducted at the same time. Finally, It can be seen through the simulation that the efficiency of the method of parallel optimization much higher than with the method of serial optimization efficiency, at the same time we also can improve the precision of the model.3. The improved algorithm is applied to predict the content of 4-CBA in PTA production process, the simulation results show the feasibility of the algorithm.