An Improved Case-based Reasoning Approach and Its Application in Soft Measurement Modeling

There are some parameters which are closely related to the quality of the product, and it is very hard to measure the parameters directly. How to get the value of these parameters is a big difficulty in control field. Usually this situation is due to the cost management or the limitations of the traditional technology. In a word, this kind of measurement is becoming a research focus in control field. And soft measurement technology is intended to solve the problem. This paper presents a new data-driven modeling approach. It uses a knowledge-based approach, that is, a Case-based reasoning approach.The most critical step in a case-based reasoning system is case retrieval. A novel technology named improved nearest neighbor method is proposed in this paper. The nearest neighbor method is widely used for case retrieval due to due to its simple principle and calculation. In this paper, all the useful data in stored in a case library and the novel case retrieval method is applied to solve new problems. There are some defects in the traditional nearest neighbor method. A new clustering algorithm and particle swarm algorithm are used to avoid these disadvantages.There are some improvements in this novel algorithm. On one hand, case library is divided into several parts because of the new clustering algorithm. Simulation results show that the new clustering algorithm improves the speed of case retrieval, at the same time, improves the accuracy of prediction. On the other hand, particle swarm algorithm is applied to select the number of the neighbors and the weights of different attributes. It shows that particle swarm algorithm is useful by simulation.At the end of this paper, the simulation to predict the esterification rate in a esterification reaction indicates that this improved case-based reasoning method is feasible and accurate.