Statistical modeling uses statistical methods to analyze the data from observation, experiment and measurement, and describes the relationship among the I/O variables.Traditional identification methods have difficulties in parameter estimation of nonlinear systems. However, neural networks have nonlinear character approximation abilities, and are advantageous in non-linear system modeling and identification. Neural network based modeling will not be restricted by nonlinear model structures, and the learning algorithm is easy to realize.In this paper, we focused on neural network based statistical data modeling, studied the characteristic of different neural networks and their applications in statistical data modeling. First we analyzed the system parameters of a sort of centrifugal fan, built its FRBFNN (fuzzy radial basis function neural network) model, and showed examples on how to guide the parameter design of centrifugal fans. Next, aiming at modeling of complex dynamic nonlinear processes, we used Elman neural network to approximate the heat exchanger, which provided a basis for heat exchanger control system design. |