共享智能櫃與眾包運輸之都會區轉運網路最佳化

Abstract

因網路購物盛行，上千萬的包裹須在城市中及時被載送至目的地，雖然雇用越多的物流人員能及時將包裹送達目的地，但物流公司須負擔高昂的人事成本，因此遞送成本較低的「眾包運送物流」近年來逐漸興起，此種配送方式為公司支付一些通勤的眾包司機，讓司機利用車內閒置空間在原路途中順道收送貨，減少公司整體的運送成本。 然而交通壅塞與顧客難以在約定時間取包裹，仍讓包裹配送成為城市物流中困難的「最初與最終一哩收送問題」。因此，結合眾包運送物流，本研究提出新型態的收送機制因應此難題：使用「智能櫃」讓顧客在方便的時間與地點取包裹，此外，智能櫃也可作為轉運站，將難以單趟直送的包裹拆成不同階段讓眾包司機接力配送，而每階段的配送皆可使用智能櫃轉運。透過眾包運送的共享效益與智能櫃的方便性與轉運功能，兩者結合的收送機制應可有效解決包裹收送問題。 本研究中，我們建構時空網路模型以解決轉運的司機匹配問題，考慮直送與轉運的可能性，以最小化整體遞送成本匹配合適的司機與包裹。然而，數學模型無法在短時間 (2小時) 內求解大型問題，因此本研究設計滾動式時窗演算法與列生成法加快求解速度，並且我們以數值實驗評估模型和算法的性能，結果顯示加入貪婪的考量部分行生成法能在短時間內求得接近最佳解的表現，我們預期可以有效幫助企業解決實際問題。 在現實生活中，即使眾包司機已經承諾協助運送包裹，也有可能無法滿足收送貨的需求。因此在第六章，我們考慮了眾包司機出現具有隨機性的問題，並設定了眾包司機協助收送貨的機率，以設計非確定性模型來解決此問題。我們使用了Expected Value of Perfect Information、Value of Sample Information以及Sample Average Analysis等方法，來評估非確定性模型的表現。

Urban growth and rapid development of e-commerce have led to urban logistics issues, which are failed delivery and delayed delivery. This study proposes an innovative distribution network taking advantage of smart lockers and shared mobility to address these problems. The smart lockers were developed to serve as a transshipping station and to make delivery more sustainable. Then, crowd-sourced drivers and courier systems are designed to relay the parcel delivery. The traditional couriers hired by the shipping company will complete the remaining unqualified shipping tasks for crowds. In order to minimize delivery costs by matching appropriate drivers with parcels, we developed a time-space network model to address the driver matching problem for parcel transportation, taking direct delivery and transshipment possibilities into consideration. We also designed a rolling horizon and a column-generation approach to solve large-scale mathematical programming problems. It is worth noting that column generation using the greedy concept can quickly identify near-optimal solutions to large-scale problems. This strategy can help businesses deal with practical challenges. Even though crowd-shippers have promised to help with delivery, they may be unable to meet the deadlines. To handle the randomness of crowd-shippers' availability, we constructed a stochastic model in Chapter 6 that includes the probability that they may not show up. To assess the stochastic model's performance, we used the Expected Value of Perfect Information, Value of Sample Information, and Sample Average Analysis.