基于三层BP网络的多指标综合评估方法及应用.pdfVIP

　1999 年 5 月 系统工程理论与实践 第 5 期　 基于三层BP 网络的多指标综合评估方法及应用 戴文战 (杭州商学院电子系, 浙江 杭州　310035) 摘要　提出一种基于人工神经网络的综合指标评估方法. 文章首先提出了一种新的归一化效用函 数, 把不同类型, 不同量纲的原始评估值转换到[ - 1, 1 ] 区间, 该效用函数较好地体现了“奖优罚劣”原 则, 同时对于神经网络又更容易学习和训练; 其次, 分析了一般多指标评估中各权值的确定方法及存 在的困难; 第三, 详细介绍了基于人工神经网络的多指标综合评价原理及实现方法, 并将之实际应用 到一柴油机行业的综合经济效益评估中, 取得了较满意的结果. 文末讨论了基于神经网络的综合评价 方法的优点. 关键词　人工神经网络　多指标综合评价　多目标决策　效用函数 A M ethod of M u ltiob jective Syn thetic Evaluation Based on A rtificial N eu ral N etw o rk s and Its A pp lication s DA IW enzhan (H angzhou Institute of Comm erce, H angzhou 310035) Abstract　In th is paper, a m ethod of m ultiobjective synthetic evaluation based on artifi cial neural netw o rk s is p resented. F irst, a new k ind of utility function, w h ich changes evaluation values w ith different dim ensions and different k inds of indexes into values in range of [ - 1, 1 ] w ith no dim ension, no t only emphasizes on the p rincip le of rew ard good and penalize bad but also m akes it easy fo r us to train neural netw o rk s to be used in th is paper. Second, how to m ake decision on the w eigh ts of indexes is discussed and . , the difficulties are analyzed T h ird the p rincip le of m ultiobjective synthetic evaluation based on artificial neural netw o rk s is described in detail, and app lied to a real synthetic 23