数据包络分析(DEA)是一种重要的数据驱动方法,可用于对一组有多个投入和多个产出的同质决策单元(DMU)进行绩效评价和改进,这些投入和产出称为绩效指标.某些绩效指标与传统DEA模型中使用的具有单个值的绩效指标不同,由于其定义或衡量标准不同,可能对应多个值,称为多值指标.绩效指标通常反映了DMU的当前生产状态,忽略了决策者的目标.为此提出了基于松弛的DEA改进模型,用于处理多值指标,可以得到帕累托最优解,并考虑了分散决策和集中决策两种常见决策场景.此外,我们通过考虑决策者目标得到扩展模型,以帮助DMU提高绩效并尽可能达到决策者的目标.基于松弛的方法和进一步考虑决策者目标增强了模型对DMU的区分能力,并为某些指标提供更符合实际的改进.通过中国长江三角洲22个城市的实例应用,说明了我们提出的模型的有效性和实用性.“,”The data envelopment analysis ( DEA) is an important data-driven method for the performance evaluation and performance improvement of a set of peer decision making units ( DMUs ) , involving multiple inputs and multiple outputs which are identified as performance indicators. However, some performance indicators, unlike conventional DEA models with one single value, may have more than one value because of different definitions or measurement standards referring to multi-valued indicators. In addition, the performance indicators reflect the current status of DMUs, which ignore the goals of decision-makers. We first propose two modified slacks-based DEA models to deal with multi-valued indicators and provide the Pareto-optimal solution in two common decision-making scenarios, namely the decentralized and centralized decision-making cases. Furthermore, we extend the models by incorporating with the goals of decision-makers to help the DMUs improve their performance and get close to the goals of decision-makers as much as possible. The slacks-based approaches and integration of goals enhance the discriminability of the models to DMUs and provide more practical improvement for some indicators. A case study of 22 cities in the Yangtze River delta region in China is used to illustrate the effectiveness and practicality of our proposed models.