Designing Fuzzy Inference System Based On Data Mining Technology

Fuzzy inference systems are composed of a set of fuzzy rules and a fuzzy inference engine. Fuzzy inference algorithms are standard and usually have little impact on the system’s performance. Therefore, the success of fuzzy modeling relies heavily on the quality of fuzzy rule base.The generation of fuzzy rules generally involves two steps:structure identification of fuzzy rules and parameter optimization of membership functions. Structure identification aims to construct a basic system with a given space partitioning and the corresponding set of fuzzy If-Then rules. The structure identification methods can be grouped into two categories:apriori knowledge-based approaches and data-driven approaches. Apriori knowledge-based approaches employ domain experts to estimate the numbers of fuzzy sets of input/output variables and then summarize the IF-THEN rules. Some successful applications show that this approach works favorably, although it is time-consuming, and some empirical studies have to be carried out before the finalization of the fuzzy rulebase. Data-driven approaches can employ data mining or other computational intelligence techniques to extract fuzzy rules from numerical data directly. Two popular and widely used fuzzy models are the Takagi-Sugeno fuzzy model and the fuzzy basis function model. It has been proved that these fuzzy models have universal approximation power as regards nonlinear maps. To identify the fuzzy basis function model, Wang and Mendel proposed a simple and practical algorithm, termed the WM algorithm, for the extraction of fuzzy rules from numerical data. This algorithm for fuzzy modeling have been highly cited and widely used by researchers and engineers in various domains due to their simplicity and effectiveness. A further study of this WM algorithm revealed that there is further opportunity to improve robustness of the fuzzy rule base.In this paper, a novel method based on data mining technique is proposed to construct fuzzy inference system. In structure identification of fuzzy rules, the conceptions of support degree and confidence degree defined in data mining are introduced to improve the fuzzy rule extraction algorithm, which make the resulting fuzzy inference system more robust with respect to the noises or outliers. In parameter optimization of membership functions, the fuzzy inference system is optimized with a partition refining strategy, adjusting the center locations of the membership functions and adding fuzzy sets, so that the structure is more suitable for the input-output data. In addition, the optimization program can select from the different structures obtained to construct a fuzzy system the one providing the best compromise between the accuracy of the approximation and the complexity of the rule set.At last, aiming at time series forecasting problems, the classical Mackey-Glass chaotic time series and the actual ship maneuvering time series is simulated. With comprehensive robustness analysis, the fuzzy inference system constructed by data mining algorithm is proved to be more robust than the system constructed by the WM method in the Mackey-Glass chaotic time series modeling. In the ship maneuvering time series modeling, our data mining algorithm is used to deal with the data information generated by an actual ship zig-zag test. The ship maneuvering model is constructed by a fuzzy inference system using fuzzy rule extracting algorithm. Compared with the traditional modeling method, the construction of a fuzzy inference system does not have to characterize the ship manoeuvrability in a unified framework, which avoids constructing a mismatched model. In addition, ship maneuvering characteristics are often interfered by the complex flow, wind and wave on the actual voyage. Lots of unavoidable noise and outliers are mixed in ship maneuvering time series records. The fuzzy inference system based on data mining technology itself is a robust system. Therefore, the system generated by data mining algorithm could meet the actual demand to make an accurate system modeling and data forecasting.