考虑专家有效信息的群决策赋权方法研究

Research on the Weighting Method for Group Decision Considering Effective Information of Experts

群决策是通过给专家分配合理的权重集成群体智慧。专家权重的分配既可以基于专家的声誉主观确定,也可以基于专家决策信息的质量客观决定。决策信息的质量一般是通过对不同专家决策信息的一致性检验来判断,一致性好的信息就是有效信息,反之则是无效信息,只有集成有效信息才能保证群决策的有效性。群决策实践中,专家对同一指标的重要性判断可能会存在一定差异性,而所有专家在不同指标上的差异性可能也是不同的,这就意味着专家给出的指标信息可能是在部分指标上是有效的,在其他指标上是无效的。因此,基于不同指标进行的一致性检验结果也会有所不同,基于某单一指标和基于所有指标的一致性检验结果一般也是不同的。现有群决策的专家权重是依据决策专家关于所有指标信息的一致性检验结果进行分配,把专家关于所有指标的信息作为整体来考虑,通过一致性检验结果剔除未通过一致性检验的专家信息,或者修正未通过一致性检验的专家决策信息,直至该专家的决策信息通过一致性检验,这种基于所有指标信息的一致性检验掩盖了专家在不同指标上决策信息的差异性,导致无法有效区分专家决策信息中的有效信息和无效信息,可能会保留通过一致性检验专家的无效信息,而删除未通过一致性检验专家的有效信息,这显然影响了群决策结果的合理性。基于此,本文提出从指标层面对决策专家的决策信息进行一致性检验,有效区分同一位专家决策信息中的有效信息与无效信息,并基于G1主观赋权方法对精确值和区间值两种信息表达下的专家权重分配方法进行了研究。首先,基于G1赋权法提出了精确值情况下的基于指标层面的一致性检验方法,确定了权重信息中有效信息和无效信息的判别标准,并建立优化模型求解最优组合权重;然后,将上述框架由精确数扩展至区间数,提出区间数下的优劣信息判断准则、一致性检验方法及权重确定的优化问题。最后,通过实际算例验证了本文所提出的群决策组合赋权方法的有效性和可行性。不同于现有研究,本文有效区分了决策专家决策信息中的有效信息和无效信息,通过指标层面进行一致性检验而不是专家层面,弥补了传统一致性检验造成的无效专家的有效信息流失和有效专家中的无效信息对结果的不利影响等问题。

Group decision-making is to integrate group intelligence by assigning reasonable weights to experts.The distribution of expert weight not only can be determined subjectively by the reputation of experts,but also can be determined objectively by the quality of expert decision-making information,which is generally judged by the consistency test.The information with good consistency is effective information,otherwise is ineffective information.The effectiveness of group decision can only be guaranteed by integrating effective information.In the practice of group decision making,experts may have differences in their judgment of the importance of the same indicator,and all experts may have differences in different indicators.The information given by experts may be effective in some indicators and ineffective in other indicators.Therefore,the consistency test results based on different indicators will be different,which means considering a single indicator or all indicators may give different consistency results.The existing method of expert weight is distributed by decision experts'consistency test result based on all indicators.The information of the experts on all indicators is considered as a whole.The expert information that fails the consistency test will be eliminated,or corrected until it passes the consistency test.The consistency test based on all indicators hides the difference of expert decision information on different indicators,so the effective information and ineffective information cannot be effectively distinguished.It may retain the ineffective information of the experts who have passed the consistency test,and delete the effective information of the experts who have not passed the consistency test,which obviously affects the rationality of the group decision results.This study is proposed to test the decision information consistency of the decision expert from the index level,effectively distinguish effective and ineffective information from the same expert's decision information.Based on the G1 subjective weighting method,the expert weight distribution method under the expression of accurate value and interval value is studied.First,based on the G1 weighting method,a consistency test method based on the index level is proposed under the condition of accurate value.The criteria for effective and ineffective information in the weight information are determined,and an optimization model is established to solve the optimal combination weight.Then,the above framework is extended from the precise number to the interval number.The criteria for good and bad information under the interval number,the consistency test method,and the optimization problem of weight determination are proposed.Finally,the effectiveness and feasibility of the proposed group decision combination weighting method are verified by an actual example.Different from the existing research,the effective and ineffective decision-making information of the decision experts is effectively distinguished.The consistency test is carried out at the index level instead of at the expert level,which makes up for the shortcoming of traditional consistency test,such as loss effective information from ineffective experts and misuse ineffective information from effective experts.