Font Size: a A A

A Study On Hybrid Case Based Reasoning Method And Its Application

Posted on:2015-11-24Degree:MasterType:Thesis
Country:ChinaCandidate:Z J CaoFull Text:PDF
GTID:2298330467486731Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Case-Based Reasoning (CBR), as an analogical reasoning method, has a very broad application prospects. It is especially suitable for fields with no complete and precise mathematical models but have extensive experience and historical records. Basic Oxygen Furnace (BOF) is a typical representative of this field. How to establish a prediction model based on CBR method, which predicts the values of endpoint carbon content and temperature according to materials and production information, is of great significance to improve production efficiency, ensure quality of steel and reduce production costs. CBR generally consists of four steps, case representation, case retrieval, case revise and case retain, and case retrieval and case revise are two important factors. In case retrieval, calculation amount is very large when calculating similarity between target and source cases. Besides that, screening of condition attributes and weight estimation are very difficult. In order to improve efficiency and validity of case retrieval, this paper introduces a method based on fuzzy clustering, mutual information and iterative learning strategies. Subtractive clustering and fuzzy C-means clustering are adapted to divide case base, and case retrieval is conducted in corresponding sub case base. Mutual information is used to estimate weights of condition attributes. And iterative learning strategies are designed to update weights and remove redundant attributes according to feedback of reasoning results. Secondly, retrieved cases cannot be completely consistent with the problems to be solved, and the solution attributes need to be properly revised. Traditional methods for case revise rely on manual adjustment, and level of automation is relatively low. A limited number of automatic revise methods can obtain good application results under certain conditions, but there are still inefficient or unsatisfactory problems. These methods don’t fully utilize the information in similar cases. According to these problems, it proposes a method based on Support Vector Machines (SVM). An error compensation model is established to adjust and modify the solutions, making use of generalization performance and robustness of SVM and characteristics of reasoning system. This model is introduced to improve effectiveness of case revise. Finally, the proposed CBR method is applied in modeling process of BOF endpoint prediction, and actual data from production records is used to test the validation of the model.
Keywords/Search Tags:Case Based Reasoning, Basic Oxygen Furnace, Case Retrieval, Case Revise
PDF Full Text Request
Related items