Rule-based metainference for crisp-type two-string fuzzy inference systems

Two-string inference systems produce output values from activations of recommending rules, as in standard fuzzy inference systems, but also with respect to inhibitions or warnings produced by negative rules. Besides assuring that a forbidden output value must not occur, the crucial point with two-string inference is how the inference results from positive and negative rules can be combined to a single output membership function by which an output value can be calculated. In this paper, a new method for combining the output membership functions of both inference strings to a combined output membership function by means of fuzzy rule-based metainference is proposed. This method generalizes existing methods and additionally provides means to adjust or combine them in a transparent way. Moreover, the possibility of designing new context-dependent or -independent metainference patterns increases flexibility and applicability of two-string fuzzy inference.