基于双层模型的置信规则库参数与结构联合优化方法

作为置信规则库优化过程的两个重要方向,参数学习和结构学习共同影响着置信规则库的建模精度和复杂度.然而,现有的置信规则库优化方法大多只关注参数学习或结构学习某一方面的研究,无法有效平衡建模精度和复杂度这对相互影响的指标.为此,本文提出了置信规则库参数与结构联合优化方法.该方法基于赤池信息准则将建模精度和复杂度两方面信息纳入统一目标,建立置信规则库联合优化目标函数;然后,建立交集假设下的置信规则库双层优化模型并提出模型求解算法;进一步拓展前提假设条件,提出并集假设条件下的置信规则库规则激活方法和权重计算方法,并提出并集假设下的置信规则库双层优化模型以及相应的求解算法.经过参数与结构联合优化之后,得到置信规则库最优决策结构.文末,引入输油管道泄漏检测案例验证所提出方法的有效性.通过与已有研究相对比,结果表明并集假设下的置信规则库联合优化方法在提高建模精度和降低复杂度方面均具有良好表现.

Bi-level model for belief rule base parameter and structure joint optimization

The parameter learning and structure learning for the belief rule base (BRB) have been the two major aspects which co-affect the modeling accuracy and the modeling complexity of BRB. So far, most BRB related studies can be categorized to either one of the two aspects. In this study, a bi-level approach of BRB parameter and structure joint optimization is proposed. First, based on the characteristics of BRB, the Akaike information criterion (AIC) has been applied to deduce a single objective which can represent both the modeling accuracy and the modeling complexity of BRB. With the AIC-based objective, a bi-level model and a corresponding algorithm for BRB joint optimization are developed under the traditional conjunctive assumption. Second, a new disjunctive assumption is proposed with new rule activation and weight calculation procedures. Furthermore, the bi-level optimization model and the optimization algorithm are updated to fit the disjunctive assumption. After BRB joint optimization, the best decision structure of BRR can be derived. Finally, the pipeline leak detection case is investigated to validate the efficiency of the proposed bi-level optimization approach. In comparison with the results of previous studies as well as that of the adaptive neural fuzzy system and the support vector machine, the BRB joint optimization approach has shown superior performance in both improving the modeling accuracy and reducing the modeling complexity.