The case study is based on the distribution system of an electrical goods wholesaler. For its operation in the South West of the UK, items need to be taken from its regional distribution centre in Avonmouth to a set of customers. The area covered includes Worcester, Swindon and Portsmouth to the east, the whole of south Wales and the south west of England to the tip of Cornwall. The operation is carried out on a daily basis Monday-Friday. The vehicles used are all 3.5 tonne GVW box vans, so there are no restrictions on the roads on which they may travel. As the items of electrical equipment are relatively small and light there are no effective constraints on the capacity of the vans. However each driver is available only for a maximum 10-h working day including the statutory breaks for driving time and working time. There are no time-window constraints for the deliveries, other than that they must all be delivered on a particular day.
Demand data were obtained for a sample of nine separate days. The number of customers served per day ranged between 40 and 64. The number of vans required is normally up to seven, though additional vans and drivers are available if required.
In order to construct the corresponding Road Timetables, data were supplied from ITIS Holdings whose Floating Vehicle Database contains speeds of vehicles on roads that have been captured through tracking devices on the vehicles. Road Timetables were constructed for each day's set of customers based on the speeds observed in 96, 15-min time bins averaged over a 3 month period in 2007. For comparison purposes, Road Timetables were also constructed using the speeds found at times of the day when the traffic was free flowing or uncongested.
In cases where the location of a customer was off the main road network covered by the ITIS data (typically on an estate or very minor road) then the time for a vehicle to transfer between the location and the main road network was estimated based on the straight line distance to a node on the network in the way described in Eglese et al (2006). This time was generally a very small proportion of the total journey time.
For each day's data, initially two runs were made using the LANTIME algorithm. The first set of runs (A) used the uncongested speeds that did not vary by time of day. The results from this correspond to what would be expected from a conventional vehicle routing and scheduling system where the speeds on each road are constant. The second set of runs (B) give the results of using the routes planned in (A), but with the varying speeds taking account of the effects of congestion at different times of day.
The results are shown in Table 1 (See PDF) .
For each of the 9 days sampled, when the routes that were constructed using constant uncongested speeds from A were used and tested using the actual time-varying speeds in B, at least one of the routes constructed became infeasible, because the total time required exceeded the 10 h allowed, sometimes by a considerable margin. These instances are indicated by bold type in the table. Over all the runs, the percentage of routes that went over time was 65% and the total extra time required to finish those routes was an average of 57 mins. In practice this may require the payment of overtime payments and could also lead to delivery problems if some deliveries are delayed beyond the normal time when customers can accept deliveries.
To overcome this problem, one strategy used by planners is to use constant speeds, but slower than the uncongested speeds to make an allowance for congestion. With a constant speed model, this will not reflect the actual variations in speed at different times of the day, but the approach might be expected to make sufficient allowance so that the actual route lengths do not exceed the 10 h allowed. Using slower speeds may lead to plans requiring more vehicle routes and drivers than strictly necessary.
In order to analyse this strategy, the algorithm was run again using constant speeds, where the original uncongested speeds were reduced by 10%. The resulting plans were then tested using the actual time-varying speeds in the same way as the previous set of runs in B. The results from these runs are shown in Table 2 (See PDF) and are labelled 'P-10%'.
The results from these runs show that even with this allowance, many of the routes planned still exceed the 600-mins time limit. The percentage of routes that went over time is 44% and the total extra time required to finish these routes is an average of 20 mins. In this case, the allowance has not been enough to provide a set of routes that are likely to be satisfactory.
Another set of runs was then carried out, again using constant speeds, but this time where the speeds were reduced by 20%. The resulting plans were then tested using the actual time-varying speeds as before. The results from these runs are shown in Table 3 (See PDF) and are labelled 'P-20%'.
A final set of runs (C) show the results from planning the routes using the LANTIME algorithm with the time-varying speed data and these are also given in Table 3 (See PDF) .
When the change in speeds is further reduced by 20% for planning, then in all instances apart from three, the routes are within the 600 mins maximum and the extra time required is only an average of 8 mins. However on each sample day, this requires the use of an additional van route compared to run set P-10%.
In contrast with the previous results, using the LANTIME algorithm with the time-varying speeds (set C) produced results where all routes were completed within the 10 h limit. As for run sets P-10% and P-20%, in many cases an additional van route was needed compared to the original plans in run sets A and B. The results from set C demonstrate that using LANTIME provides a more reliable basis for planning routes in terms of the time needed to complete each route.
Table 3 (See PDF) also presents an estimate of the CO2 emissions for each of the run sets. These have been calculated using the speed along each road in the route using the emissions function provided in the National Atmospheric Emissions Inventory. This can be accessed online (at www.naei.org.uk). For this case study, the figures used are for Euro II Diesel LGVs. The tables provided allow an estimate to be made of various emission factors in terms of emissions per kilometre for different average speeds. They may not fully reflect actual emissions that may be affected by the amount of irregularity in speed, weight of load, road inclines and other factors. For this case, as the customer orders are relatively light compared to the weight of the van, no attempt has been made to modify the function for the weight of goods carried at each stage of the route. A good discussion of the issues involved in the estimation of CO 2 emissions from road freight transport can be found in McKinnon and Piecyk (2009). The evaluations could have been made for other harmful pollutants, but only CO2 emissions have been evaluated as a major contributor to the greenhouse effect. All the calculations have been made in grams and then rounded in the results to the nearest kilogram.
When the total emissions per day are compared for run sets P-20% and C, the total emissions for run set C were usually lower than those for run set P-20%, though not in every case.
Table 4 (See PDF) summarises the total distance, total time required and the total CO2 emissions for run sets P-20% and C. It shows that the total distance travelled and the total time required for run set C were less than those for P-20%. The reduction is about 7% when compared with the P-20% run set. This is because the LANTIME algorithm using the time-varying speed data tends to avoid routes where congestion is high, speeds are low and CO 2 emissions are relatively high. By searching for the fastest routes, it tends to avoid congestion and only uses longer routes when the vehicles can travel faster at a speed closer to the optimum for emissions per kilometre.