Preference Functions (PF) are actually selected based on the practical nature of the attributes. To know how it's done in practice, you may refer to the paper where PROMETHEE was proposed first time. Also PROMETHEE doesn't restrict to stay only within 6 types as you have the freedom to define your own PF if you are not satisfied with the given list.
Actually, different types of preference functions are devised to be used in specific situations where they can serve better. Generally, different types are as follows: (1) usual criterion which is used when the decision maker cannot allocate importance for the differences between criteria values and only seems to know the formula “the more the better”; (2) U-shape function which is suitable for strict comparison of any two alternative (among two criteria the one with better value has complete preference which is 1, while the other one is assigned 0 preference), (3) V-shape/linear function which accounts for moderate comparison and in contrary with U-shape function it establishes a linear correlation between the point of indifference, 0, and the point of strict preference, 1, (4) level criterion, (5) V-shape with indifference preference, and (6) Gaussian criterion.