Inbound operations in distribution centers face high uncertain truck arrival times. Since unloading relies on human labor, setting the number of workers before knowing actual arrivals can result in mismatches: understaffing (leading to delays and overtime) or overstaffing (increasing idle costs). This thesis tackles this problem by proposing a stochastic optimization model that jointly optimizes the number of workers per shift and schedules inbound trucks considering arrival uncertainty. A two-stage stochastic model is proposed. In the first stage, the number of regular workers is set before truck arrival times are known. In the second stage, once arrivals are revealed, the model optimizes the unloading schedule. It accounts for overtime costs, when regular staff cannot complete tasks in shift hours and waiting penalties if trucks are served late. The objective is to minimize regular labor, expected overtime, and waiting costs. We define a risk-neutral (RN) model minimizing expected costs, and a risk-averse (RA) variant using CVaR for extreme scenarios. We also propose two benchmarks, a non accounting for uncertainty (NAU) model, assuming arrivals follow the plan, and a lower bound (LB) with perfect knowledge of the actual arrival times. The problem is solved using sample average approximation and ternary search to find optimal solutions in tractable times. Data from a postal distribution center is used to generate instances. Results show that accounting for uncertainty brings the RN solution within 0.2% of the LB, while the NAU differs by 1.7%. It also leads to hiring more workers (+12%), allowing to reduce overtime costs by 55.4% compared to NAU, and finishing operations 1.5 hours earlier. The gap widens under a first-come-first-served rule: NAU is 24% above the LB, while RN is only 17%, highlighting how reactive dispatching amplifies the cost of ignoring uncertainty. The RA model improves costs under worst-case scenarios, reducing the CVaR value by 0.37% with only a 0.09% increase in cost under extreme congestion. A sensitivity analysis shows that scheduling arrivals earlier in the shift further reduces total costs by 1.33% compared to uniform scheduling with RN.
Le operazioni in ingresso nei centri di distribuzione affrontano un’elevata incertezza negli orari di arrivo dei camion. Poiché lo scarico dipende dalla manodopera, definire il numero di lavoratori prima di conoscere gli arrivi effettivi può causare inefficienze: carenze di personale generano ritardi e straordinari, mentre un’eccedenza aumenta i costi per risorse inutilizzate. La tesi propone un modello di ottimizzazione stocastica che determina congiuntamente il numero di lavoratori per turno e il programma di scarico, tenendo conto dell’incertezza negli arrivi. Il modello è a due stadi: nel primo si decide il numero di lavoratori regolari; nel secondo, noti gli arrivi, si ottimizza lo scheduling. Sono considerati i costi del lavoro regolare, gli straordinari attesi quando il personale è insufficiente e le penalità per ritardi nel servizio. Si analizza un modello risk-neutral (RN), che minimizza il costo atteso, e una variante risk-averse (RA), basata sulla conditional value-at-risk (CVaR), per scenari estremi. Sono inoltre introdotti due benchmark: un modello non accounting for uncertainty (NAU), che assume arrivi pianificati, e un limite inferiore (LB), con piena informazione. La risoluzione avviene tramite sample average approximation e ricerca ternaria. I dati derivano da un centro di distribuzione postale. I risultati mostrano che RN si avvicina allo 0.2% al LB, mentre NAU devia dell’1.7%. RN richiede il 12% di personale in più, riduce del 55.4% i costi per straordinari rispetto a NAU e anticipa la fine delle operazioni di 1.5 ore. Con una regola first-come-first-served, il gap si amplia: NAU è sopra il LB del 24%, RN del 17%. RA migliora le prestazioni nei casi peggiori, riducendo CVaR dello 0.37% con un aumento limitato del costo totale (+0.09%). Un’analisi di sensibilità mostra che anticipare gli arrivi all’inizio del turno riduce i costi complessivi dell’1.33% rispetto a una pianificazione uniforme con RN.
Stochastic workforce sizing under truck arrival uncertainty in distribution centers
CALDERON ROMERO, BARBARA CATALINA
2024/2025
Abstract
Inbound operations in distribution centers face high uncertain truck arrival times. Since unloading relies on human labor, setting the number of workers before knowing actual arrivals can result in mismatches: understaffing (leading to delays and overtime) or overstaffing (increasing idle costs). This thesis tackles this problem by proposing a stochastic optimization model that jointly optimizes the number of workers per shift and schedules inbound trucks considering arrival uncertainty. A two-stage stochastic model is proposed. In the first stage, the number of regular workers is set before truck arrival times are known. In the second stage, once arrivals are revealed, the model optimizes the unloading schedule. It accounts for overtime costs, when regular staff cannot complete tasks in shift hours and waiting penalties if trucks are served late. The objective is to minimize regular labor, expected overtime, and waiting costs. We define a risk-neutral (RN) model minimizing expected costs, and a risk-averse (RA) variant using CVaR for extreme scenarios. We also propose two benchmarks, a non accounting for uncertainty (NAU) model, assuming arrivals follow the plan, and a lower bound (LB) with perfect knowledge of the actual arrival times. The problem is solved using sample average approximation and ternary search to find optimal solutions in tractable times. Data from a postal distribution center is used to generate instances. Results show that accounting for uncertainty brings the RN solution within 0.2% of the LB, while the NAU differs by 1.7%. It also leads to hiring more workers (+12%), allowing to reduce overtime costs by 55.4% compared to NAU, and finishing operations 1.5 hours earlier. The gap widens under a first-come-first-served rule: NAU is 24% above the LB, while RN is only 17%, highlighting how reactive dispatching amplifies the cost of ignoring uncertainty. The RA model improves costs under worst-case scenarios, reducing the CVaR value by 0.37% with only a 0.09% increase in cost under extreme congestion. A sensitivity analysis shows that scheduling arrivals earlier in the shift further reduces total costs by 1.33% compared to uniform scheduling with RN.File | Dimensione | Formato | |
---|---|---|---|
2025_07_Calderón_Executive Summary_02.pdf.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
924.86 kB
Formato
Adobe PDF
|
924.86 kB | Adobe PDF | Visualizza/Apri |
2025_07_Calderón_Thesis_01.pdf.pdf
accessibile in internet per tutti
Descrizione: Thesis
Dimensione
5.7 MB
Formato
Adobe PDF
|
5.7 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/240516