Temperature-controlled power normalization of belief function

Belief function describes uncertainty in Dempster–Shafer theory. However, when conflicts arise between different information sources, the fusion results may be counterintuitive and unreliable. The traditional conflict handling method and the discounting method do not adapt well to dynamic changes. Temperature is an important parameter influencing the smoothness and concentration of probability distributions. Based on its characteristics, we propose a temperature-controlled power normalization of belief function. The temperature parameter can change the support assigned to subsets and reshape the resulting distribution. It also changes the entropy and thereby affects the level of uncertainty. When the temperature threshold is reached, the belief distribution undergoes a phase transition, which is reflected in changes in the ranking of subsets and in the identity of the most supported subset. Examples illustrate the trend of temperature influencing belief distributions. The application uses the proposed method to solve conflicts in data fusion and compares it with common fusion methods. The proposed method provides a parametric way to adjust mass assignments under conflict or high uncertainty, and offers a flexible framework for uncertainty regulation in Dempster–Shafer theory.