An Improved Case-Based Reasoning Method and Its Application in Endpoint Prediction of Basic Oxygen Furnace

Min Han,Zhanji Cao

DOI: https://doi.org/10.1016/j.neucom.2014.09.003

IF: 6

2015-01-01

Neurocomputing

Abstract:Case retrieval and case revise (reuse) are core parts of case-based reasoning (CBR). According to the problems that weights of condition attributes are difficult to evaluate in case retrieval, and there are few effective strategies for case revise, this paper introduces an improved case-based reasoning method based on fuzzy c-means clustering (FCM), mutual information and support vector machine (SVM). Fuzzy c-means clustering is used to divide case base to improve efficiency of the algorithm. In the case retrieval process, mutual information is introduced to calculate weights of each condition attribute and evaluate their contributions to reasoning results accurately. Considering the good ability of the support vector machine for dealing with limited samples, it is adopted to build an optical regression model for case revise. The proposed method is applied in endpoint prediction of Basic Oxygen Furnace (BOF), and simulation experiments based on a set of actual production data from a 180t steelmaking furnace show that the model based on improved CBR achieves high prediction accuracy and good robustness.