Immune Agent-Based Neural Networks Soft-Sensor and its Application

Aiming at difficulty modeling of large amounts of industrial process data, a novel soft-sensor based on artificial immune multiagent and multiple model Radial Basis Function(RBF) networks is proposed. The method is to predict the qualities of manufactred products from process variables. Where, artificial immune T-cell and B-cell agents with different tasks of clustering work cooperatively to accomplish the common goal of soft-sensor model training. Immune memory and pattern recognition provide high efficiency of predicting. Multiple model technique is introduced to improve the computation and performance of soft-sensor. The prediction of dry point of naphtha produced in a practical industrial process is carried out as a case study. Results obtained indicate that proposed method provides oil quality prediction with high efficiency and accuracy which is capable of learning the relationship between process variables measured during the production.