The Impact of Different Approximate Reasoning Methods on Fuzzy Signal Controllers

1 INTRODUCTION The use of fuzzy signal controllers has been proven to be quite effective. Their performance in comparison with traditional pretimed or vehicle-actuated control modes has been assessed by different authors (Niittymäki and Nevala, 2001). In all cases, the fuzzy logic controllers provide better traffic operation according to the usually adopted measures of traffic performance such as delay and number of stops. An analysis of the theoretical aspects of the decision-making logic of some of these controllers has shown that union operator is generally used for implementing the connective " also " and the intersection operator for the connective " or " (Jacques et al., 2002). It also shows that in most cases the fuzzy relation among fuzzy sets derived from the fuzzy condition proposition (or statement) is the minimum operation rule of fuzzy implication (Rc). In controllers where another relation has been used, as is the case with the product operation rule (Rp), no discussion is presented regarding the impact of this choice on the controller's response. The same was observed with respect to defuzzification strategies adopted. Braae and Rutherford (1978) have drawn attention to the fact that the control element in generic fuzzy logic controllers " is dependent not only on the rules and fuzzy sets assumed, but also on the mathematical operations chosen to define composition, classification and interpretation ". In light of this, this research aims to assess the impact of using alternative operators, fuzzy relations and defuzzification strategies on the performance of the signal controllers' control action. Initially, the controllers' responses to different decision-making logic and to defuzzification interface are analysed based on results provided by MATLAB. Afterwards, the impact of these controllers' responses on traffic performance is assessed with the aid of the simulation program HUTSIM (Kosonen, 1999). Aspects such as fuzzification interface and knowledge base definition are outside the scope of this work.