考虑投入产出的常规公交调度绩效回溯性动态评价

Retrospective Dynamic Evaluation of Bus Scheduling Performance Considering Input-Output Relationship

摘要:

实时调度因其灵活性在公交运营组织中发挥着重要决策作用。现行调度（员）绩效评价虽然能够考核行车计划执行情况，却忽略了在乘客体验和企业投入产出层面对调度效果考查，也不支持回溯过往调度实施效果。为使调度员能够精准提升调度技能和绩效、解决调度决策滞后造成的运能浪费等问题并提高乘客满意度，从投入产出及回溯性动态评价2个维度提出基于三阶段Super-SBM（super-efficiency slack-based measurement）模型的常规公交调度绩效评价方法，考虑路况及客流对调度效果的影响，在现行指标基础上从乘客角度增加运营速度、候车时间变化率及满载率标准差3项产出指标，构建评价体系。对2条上海公交线路的实际调度进行回溯性评价，验证了本文评价方法能准确区分调度效果差异。结果表明，高峰时段灵活的调度策略比在平峰能产生更明显的效益，且调度策略在不同时空场景呈现差异化适用性。通过回溯分析历史调度记录的绩效，一方面可以针对各场景为调度员提供调度决策支持，另一方面可为基于机器学习的公交实时调度策略生成技术挑选合适的学习样本，从而提升常规公交运营组织效率。

Abstract:

Real-time scheduling plays an important decision-making role in public transport operation organization because of its flexibility. Although the current scheduling （or dispatcher） performance evaluation can assess the implementation of the bus schedule， it ignores the passenger experience and enterprise input-output evaluation of the scheduling performance. It also does not support backtracking the implementation performance of the scheduling in the past. In order to enable dispatchers to accurately improve their dispatching skills and performance， solve problems such as wasted capacity caused by lagging dispatching decisions and increase passenger satisfaction， from the two dimensions of input-output and retrospective dynamic evaluation， this paper proposes a performance evaluation method for bus scheduling based on the three-stage Super-SBM （super-efficiency slack-based measurement） model. Considering the impact of road conditions and passenger flow on the scheduling performance， and based on the existing indicators， three output indicators， i.e.， travel speed， waiting time change rate， and standard deviation of load factor， are added from the perspective of passengers to build an evaluation system. The retrospective evaluation of the actual scheduling of two bus routes in Shanghai verifies that the evaluation method proposed in this paper can more accurately distinguish the difference in scheduling performances. The results show that the flexible scheduling strategy in peak periods can produce more obvious benefits than that in off-peak periods， and the scheduling strategy shows different applicability in different space-time scenarios. The retrospective analysis of the performance of historical scheduling records， on the one hand， can provide scheduling decision support for dispatchers in different scenarios， and on the other hand， can select appropriate learning samples for the real-time bus scheduling strategy generation technology based on machine learning， to improve the efficiency of bus operation and organization.