Online decision making and automatic decision model adaptation☆

https://doi.org/10.1016/j.cor.2008.04.009Get rights and content

Abstract

The paper investigates an online version of the vehicle routing problem with time windows, in which additionally arriving requests cause the revision of so far followed routes and schedules. An extended online optimization framework is proposed, which automatically adapts to problem variations and enables the explicit consideration of up-to-date knowledge about the current performance of the controlled system. Actually, we use the mean punctuality observed in the transportation system to adjust the objective function utilized for solving the next decision problem instance. The search is guided toward least cost solutions coming along with high punctuality. We prove the applicability of this approach within comprehensive numerical experiments.

Introduction

The supply chain of a product describes the sequence of activities to be carried out in order to create the desired output from one or several inputs factors. Supply chain planning (SCP) aims at achieving the highest possible efficiency of a supply chain by coordinating and consolidating the necessary material flows (“processes”) so that economies of scale are exploited to the largest possible extent.

Recent trends in the management of supply chains compromise the successful application of existing concepts for computational planning support.

  • •

    The ability to consider unexpected events in an ad hoc fashion is propagated as significant competitive advantage. The continuous incorporation of recent problem data requires a continuous plan revision.

  • •

    The partners forming the supply chain are not willing to give up the responsibility and self-reliance for the material flow decisions in their part of a supply chain. Consequently, the centralized supply chain wide top-down material flow determination and the goals of the incorporated partners are sometimes contradicting.

Contracts between supply chain partners are fixed for several months and must consider both the responsiveness of the involved partners to dynamics (e.g. demand variations and peaks) and the partners’ autonomy in the operational deployment planning. Although a supply chain is built by independent partners, one of them, the supply chain coordinator is dedicated and entitled to persuade the independent partners to behave and act in the sense of the superior supply chain goals instead of the subordinate partner's aims.

Computer-supported decision making is necessary for all supply chain partners. The definition of a suitable mathematical decision model is a prerequisite for the successful application of automatic decision making tools like optimization algorithms. However, the fine-tuning of such a model is a sophisticated task that typically requires some trial-and-error runs in order to identify the best parameter setting. Solving a concatenated sequence of decision problem instances is referred to as online optimization. The definition and the solving of a new instance are triggered by events that compromise the realization of the so far optimal solution. There is no time to experiment on the right parameters for the decision model of the new problem. Here, the right parameters have to be adjusted automatically.

Within this article, we investigate the impacts of different configurations for the interaction between the supply chain coordinator and a transport-providing partner in a given supply chain. We analyze the implications of different intervention rights that enable the coordinator to bias the planning decisions of the transport partner by adjusting relevant decision model parameters. We show that a performance-oriented adaptation of the transport partner's decision logic has positive impacts on the overall supply chain reliability. An extension of the well-established online decision making framework is proposed. It enables a planning system of the coordinator to detect supply chain performance variations and to implement autonomously the necessary decision model adaptations in the planning system of the subordinate transport partner. Numerical simulation experiments are reported in order to assess the proposed extension.

Section 2 introduces the investigated problem. Section 3 reports the results of numerical simulations using the pure online decision making framework without performance feedback. Section 4 presents the concept for automatic decision model adaptation. Section 5 reports results of comprehensive numerical experiments.

Section snippets

Vehicle deployment planning problem with uncertain demand

We introduce the investigated supply chain setting in this section. Related scientific literature is compiled in Section 2.1. The supply chain layout is described in Section 2.2. Three configurations for the interaction between the coordinator and a transport partner are described and motivated in Section 2.3. A model for the deployment problem of the transport partner is presented in Section 2.4. The derivation of artificial testcases used in the simulation experiments is explained in Section

Online planning in HARD- and PEN-configurations

This section reports the experimental setting and findings for the HARD- and the PEN-configurations. In Section 3.1, we describe the configuration of a hybrid search algorithm for solving the instances TP0,TP1,…. Section 3.2 compares techniques for ensuring the considerations of the constraints (2), (3), (4), (5), (6), (7) during the solving of SP(ti). Section 3.3 describes the experimental setup. The achieved results are presented and discussed in Section 3.4.

Image modification approaches

PEN-configured as well as HARD-configured supply chain settings produce transportation plans causing nearly the same costs. Nevertheless, the two configurations show a noticeable difference in the quality of the generated transportation plans if the tariff level α climbs up. In this case a HARD-configuration performs significantly better with respect to the reliability of request fulfillment. First, the HARD-configuration produces less severe punctuality decreases (Table 1). Second, the

Computational experiments

The SDAD-configuration prevents long-lasting phases of poor punctuality of the logistic system even if the tariff level for subcontraction is very high (α=3). As shown in Fig. 5, it keeps the maximal number of waiting requests as low as observed for the HARD-configuration. Furthermore, the time required to dismantle the enlarged queue of waiting requests is the same for SDAD- and HARD-configurations.

In order to analyze the impacts of the objective function adaptation (realized by the adaptation

Conclusions and future research

We have investigated the automatic decision support for a dynamic decision problem. From the observed results we conclude that it is necessary to adapt the used formal decision model if the decision making situation has changed and if the used decision logic is not suitable anymore to fulfill the superior supply chain wide planning goals. This is observed in the α>1-cases where the punctuality in workload peak situations cannot be maintained using the static PEN-configuration. We have shown

References (26)

  • D. Wasserburger

    Dynamische Tourenoptimierung durch verkehrstelematische Daten

  • M. Arnold et al.

    A survey of adaptive optimization in virtual machines

    Proceedings of the IEEE

    (2005)

    • Cited by (0)

    ☆

    This research was supported by the German Research Foundation (DFG) as part of the Collaborative Research Centre 637 “Autonomous Cooperating Logistic Processes” (Subproject B7).

    View full text
    Copyright © 2008 Elsevier Ltd. All rights reserved.