Research and Application of Function Linked Neural Network Based on Error Compensation

In the process of chemical production, some key process variables need be measured accurately. Soft sensor is of great importance. Due to the increasing difficulties of modern processes, it is harder and harder to develop accurate soft sensor. FLNN (Function linked Neural network is a promising model for building soft sensor. The industrial data tend to be high-dimensional and highly nonlinear. The accuracy of the developed soft sensor cannot meet the requirement by the traditional FLNN. To solve this problem, this paper put forward a novel model to improve the accuracy of the FLNN. The main idea of this proposed method is error compensation. The proposed method is called as error compensation based FLNN (EC-FLNN) where two FLNN models are established. The first one is to predict the output value, and the second one is to predict the error value. The sum of the results of the two models is obtained as the final outputs. The UCI standard data set and PTA production data set are used to verify the performance of the proposed method. Simulation result shows that EC-FLNN has better precision than the traditional FLNN.