220 likes | 360 Views
V ehicle routing using remote asset monitoring: a case study with Oxfam. Fraser McLeod , Tom Cherrett (Transport) Güneş Erdoğan , Tolga Bektas (Management) . OR54, Edinburgh, 4-6 Sept 2012. Background. www.oxfam.org.uk/shop. Donation banks. Oxfam bank sites in England. Case study area.
E N D
Vehicle routing using remote asset monitoring: a case study with Oxfam Fraser McLeod, Tom Cherrett (Transport) GüneşErdoğan, Tolga Bektas (Management) OR54, Edinburgh, 4-6 Sept 2012
Background www.oxfam.org.uk/shop
Donation banks Oxfam bank sites in England
Problem summary (requirements) Visit shops on fixed days Visit banks before they become full Routes required Monday to Friday each week Start/end vehicle depot Single trips each day (i.e. no drop-offs)
Problem summary (constraints) • Heterogeneous vehicle fleet • 1 x 1400kg (transit van) • 3 x 2500kg (7.5T lorry) • Driving/working time constraints • Time windows for shops
Objectives • Maximise profit (£X per kg – £1.50 per mile) • where X = f(site) (e.g. 80p/kg from banks; 50p/kg from shops) • Avoid banks overfilling • prevents further donations (= lost profit) • upsets site owners • health and safety
Data (locations, time, distance) • Postcodes for 88 sites: • 1 depot • 37 bank sites • 50 shops • Driving distances/times between 3828 (= 88x87/2) pairs of postcodes • Commercial software • Times calibrated using recorded driving times
Data (demand) Weights collected from shops and banks (April 2011 to May 2012) Remote monitoring data (from July 2012) Shop demand = average accumulation rate x no. of days since last collection Bank demand – randomly generated
Assumptions (bank demand) • Demand at bank i, day j = Xi,j =max(Xi,j-1 + di,j-1, bank capacity) where d=donations = Yi,j.Zi,j Y = Bernoulli (P = probability of donation) Z = N(m, s) = amount donated • m= mean daily donation amount, excluding days where no donations are made • sestimated from collection data • bounded by [0, bank capacity]
Assumptions (collection time) Collection time = f(site, weight) = ai + bi xi
Solution approach • Look ahead period = 1 day (tomorrow) • Minimum percentage level to be collected • (50% and 70% considered) • Overfilling penalty (applied to banks not collected from) • fill limit (%) (75% and 95% considered) • financial penalty (£/kg) (£10/kg considered)
Solution approach • Tabu search • Step 1 (Initialization) • Step 2 (Stopping condition): iteration limit • Step 3 (Local search):addition, removal and swap • Step 4 (Best solution update) • Step 5 (Tabu list update) • Go to Step 2
Results / KPIs • 20 consecutive working days • 3 random starting seeds • Performance indicators • # bank visits • profit • distance • time • weight collected and lost donations
Probability of donation Results (# bank visits) Penalty filllevel
Conclusions & Discussion Bank visits could be substantially reduced But benefits are limited by the requirement to keep shop collections fixed Can we improve our modelling approach?