| 摘要 |
PURPOSE:To attain easy handling for a fuzzy model by generating consequent part membership functions with the same centroid and area leaving centroid obtained by a centroid calculation part and an area obtained by an area calculation part as it is. CONSTITUTION:The area and the centroid of the consequent part membership function in the fuzzy model are found by the area calculation part 18c and the centroid calculation part 18b. When the fuzzy model is realized in a neural network, the fuzzy model is learned at a neural network learning device, and the area and the centroid of the consequent part membership function after obtained learning are found. Thence, a new membership function with the same area and the same centroid is generated by a consequent part membership function shaping part 18d. The new membership function is changed to a function for which evaluation by a person can be easily rendered, for example, the one representing a triangle, especially, an isosceles triangle, a right-angled triangle, or a trapezoid. Since the area and the centroid are left as it is even when such shaping is applied, no influence is given to the final conclusion, which facilitates only the evaluation of the consequent part membership function. |
