WO2015154831A1

WO2015154831A1 - Dynamic fleet routing

Info

Publication number: WO2015154831A1
Application number: PCT/EP2014/075260
Authority: WO
Inventors: Nitin MASLEKAR; Konstantinos GKIOTSALITIS
Original assignee: Nec Europe Ltd.
Priority date: 2014-04-07
Filing date: 2014-11-21
Publication date: 2015-10-15
Also published as: JP2015201006A; JP5944431B2

Abstract

A method of dynamic fleet routing, wherein said fleet includes a number of vehicles that perform delivery services, and wherein an initial delivery plan is generated based on a customer demand for deliveries, wherein said initial delivery plan specifies routes for said vehicles and schedules customers of the delivery services within the routes, is characterized in the steps of splitting the planning period of said initial delivery plan into a number of time slots, associating each of said time slots with a slot inconsistency factor depending on factors that contribute to dynamicity within the respective time slot, and based on said slot inconsistency factor, allotting to each of said time slots an individual slack time to accommodate any dynamic changes. Furthermore, the invention relates to a corresponding fleet management system.

Description

DYNAMIC FLEET ROUTING

The present invention relates to a method of dynannic fleet routing and to a fleet management system, wherein said fleet includes a number of vehicles that perform delivery services, and wherein an initial delivery plan is generated based on a customer demand for deliveries, wherein said initial delivery plan specifies routes for said vehicles and schedules customers of the delivery services within the routes. With the recent growth in online shopping, logistic firms are required to improve their planning process in order to accommodate more challenging customer behavior and external conditions which involves a higher level of dynamicity. The challenge here is not only to ensure the delivery of goods, but also the speed and quality of delivery (i.e., via reducing the time period that a customer waits at home for a delivery).

Many logistic companies plan a daily 12-hour delivery window. These solutions are two-step processes (plan-execute). As schematically illustrated in Fig. 1 , many dynamic factors like customer behavior (for instance, missed deliveries, new demands, recipient's time and location changes, etc.) and unpredictable external conditions (e.g. traffic congestion, weather, or the like) affect this planning which results in many anomalies. In particular, these anomalies may include cases in which the customer is inconvenienced, the logistic companies incur additional costs, the daily delivery plan of the logistic company is affected, or in which there are wider environmental impacts, owing to additional vehicle trips. In this context, logistic companies are under pressure to meet the expected QoE (Quality of Experience) demands of customers with their existing fleet infrastructure along with minimizing the operational costs. Generally, logistic problems as outlined above belong to the domain of vehicle routing problems, commonly known in the art as VRP. In vehicle routing problems a significant potential for research is concentrated in dynamic planning of delivery routes, especially dynamic re-scheduling and re-routing of vehicles. The existing dynamic systems consider conditions which include real-time variations in travel times between demand nodes and real-time service requests. Demands can arrive at any time during a planning period. For example, in the Dynamic Vehicle Routing Problem (DVRP), new customer orders appear over time, and new routes must be calculated while the existing solution is being executed. In the literature, many methods and strategies have been proposed to tackle DVRPs. Generally, DVRP is considered as the extension to the standard VRP by decomposing a DVRP as a sequence of static VRPs and then solving them with, e.g., ant colony system algorithm. Some of the known solutions use a reactive method (agent-based constraint programming) to solve DVRPs, while some solutions introduce a consensus approach to the problem.

However, it is common to all prior art solutions that they do not address the fact that the current planning horizon might involve dynamicity which affects the plan and hence the operational cost. In such cases, if a typical VRP approach is used for re-planning (i.e. re-planning the whole schedule from scratch), many vehicle schedules and pre-arranged delivery times may be affected, thus causing significant performance inefficiencies (high overhead, nervousness, errors, and high costs). Along with the operational costs for the operator, the re-planning may affect the QoE of the customers which, later on, might affect the business of the fleet operators.

In view of the above it is an objective of the present invention to develop a method of dynamic fleet routing and a fleet management system of the initially described type in such a way that a dynamic and intelligent planning of the delivery schedule is enabled that takes into account dynamic changes without incurring extensive additional operational costs or customers' inconvenience (i.e., changing the delivery times of several customers).

In accordance with the invention, the aforementioned object is accomplished by a method comprising the features of claim 1. According to this claim such a method is characterized in the steps of

splitting the planning period of said initial delivery plan into a number of time slots, associating each of said time slots with a slot inconsistency factor depending on factors that contribute to dynamicity within the respective time slot, and

based on said slot inconsistency factor, allotting to each of said time slots an individual slack time to accommodate any dynamic changes.

Furthermore, the above mentioned objective is accomplished by a fleet management system comprising the features of claim 15. According to this claim such a system is characterized in that it comprises computation means that are configured

to split the planning period of said initial delivery plan into a number of time slots,

to associate each of said time slots with a slot inconsistency factor depending on factors that contribute to dynamicity within the respective time slot, and

based on said slot inconsistency factor, to allot to each of said time slots an individual slack time to accommodate any dynamic changes.

According to the present invention it has been recognized that in fleet routing dynamic changes can be taken into account without incurring extensive additional operational costs by dividing the operational time in time slots, by determining an inconsistency related to each time slot and by introducing appropriate buffer times. Insofar, the proposed invention provides a solution to analyze multi-dimensional dynamicity which affects the plan of delivery services and can thus be regarded as an inconsistency based dynamic and intelligent fleet routing for spatio-temporal demand variations. According to specific embodiments, the invention involves determining inconsistency in the route due to customer behavior/external factors, assign an inconsistency-based buffer time to each route and schedule the customers within the route based on the determined inconsistency.

The system according to the present invention can be either stand alone or can be integrated into an existing planning tool for effective planning and re-scheduling of deliveries. The invention provides a strategy to counter and respond to varying dimensions of dynamicity which effects the planning in order to optimize fleet operators OPEX (Operational Expenditure) and customer QoE.

According to a preferred embodiment the initial delivery plan may be generated based on the demand for deliveries by applying a VRP (Vehicle Routing Problem) algorithm. For this purpose, any known VRP algorithm may be used.

As mentioned above, according to the invention the planning horizon of the base VRP solution is split into smaller time slots. According to preferred embodiments two different time slot schemes may be used to split the time horizon. A first scheme may split the horizon into evenly distributed, customer-oriented time slots of equal lengths. The time frame of those time slots is kept as minimum as possible (e.g. 1 hour) in order to increase the QoE since customers will have to wait less for their delivery. A second scheme may be operational-oriented and may split the time horizon proportionally to the number of customers in close proximity (clustered customers). The exact definitions of 'proximity' are implementation specific and may be configured by the users of the system, e.g. a logistic company. Apart from geographic proximity, customers may be clustered into groups based on any other factor like low tonnage, ability to navigate in small streets/inner cities, etc. In any case, cluster customers are allotted in proportional time slots and, after that, time slacks are allotted based on the inconsistency of each time slot. As will be explained in more detail below, the time slot inconsistency may be derived from log analysis on the customers' profiles (i.e., missed delivery records of each customer), log analysis on previous travel times and inference of traffic conditions and the number of customers allocated at the time slot - together with their geographic proximity.

According to a further preferred embodiment it may be provided that the management fleet system can interact with the customers, i.e. via appropriate interfaces, where customer can be offered a number of time slots from a pre- calculated list and will be requested to provide their preference for the deliveries: E.g., we will deliver between 9 and noon, please indicate the preference (by sms) with the following syntax: 9-1 O>1 1 -12>10-1. Time slots may be assigned in a way that the end of one time slot overlaps with the beginning of the next. In preferred embodiments the factors that contribute to dynamicity within a time slot and that are taken into account for determining the inconsistency factor include the customers' reliability. Advantageously, a customer's reliability is derived from log-analysis of recorded customer profiles, in particular by taking into consideration the occurrence of missed deliveries in the past. To this end, the fleet management system may comprise a log analyzer for analyzing log information on customer's profiles, in particular with customers' missed delivery records, on the total travel times spent for accommodating a number of deliveries within a geographical area and for different traffic conditions, on traffic conditions in different geographic area during different day times, and/or on the behavior of vehicle drivers in reaction to dynamicity. The log information may be transmitted from the vehicles to the fleet management system via appropriate communication means, which may also be used to transmit instructions regarding new and/or adapted routes from the fleet management system to the vehicles of the fleet on the road.

Based on the outcome of a customer profile analysis, it may be provided that a customer is assigned a quasi-static behavior index in case his reliability exceeds a predefined configurable threshold. In order to minimize the effect of dynamism on the operational costs, customers with higher inconsistency may be scheduled at the beginning of time slots, for instance by assigning appropriate weight factors.

According to further preferred embodiments, the factors that contribute to dynamicity within a time slot and that are taken into account for determining the inconsistency factor include at least one of the factors:

the estimated time for serving a number of deliveries within the time slot, the expected traffic at the examined geographic area during the time period contained in the time slot, and

the reaction of drivers of the vehicles on occurred dynamicity during the execution of initial delivery plan.

As mentioned above, at the end of each time slot, time slacks are introduced to accommodate any dynamic changes over the planned horizon. Advantageously, to optimize the operational costs, the lengths of the individual slack times allotted to the time slots may be designed to be the longer the higher is the inconsistency which exists in the slot. In particular, it may be provided that the length of a time slack is designed to be proportional to the inconsistency which exists in the respective time slot.

With respect to an effective and reliable calculation of the quality of experience (QoE) for the customers it may be provided that the QoE is determined as a function of the percentage of deliveries executed in the scheduled time slot, the number of customers who received their deliveries within the pre-arranged delivery time slot, and a customer retention index.

By using time slots and time slacks, spatio-temporal changes on delivery demand are served without increasing significantly the base operational cost and disruptions on the deliveries of other customers are kept to a minimum. During the execution, it may be provided that the time slotted solution is monitored continuously for any missed deliveries or spatio-temporal demand changes. To ensure the delivery, a reassignment method may be applied via executing an open VRP algorithm recursively for all the available operational-oriented time slots such that the first and the last customer of each time slot remain always the same. After applying the open VRP algorithm for all available time slots, the most appropriate operational-oriented time slot for the attempted redelivery is selected. The criterion for the time slot selection may be based on the minimization of the incurred operational cost. However, as will be appreciated by those skilled in the art, different optimization codes may be specified by the system operator.

There are several ways how to design and further develop the teaching of the present invention in an advantageous way. To this end it is to be referred to the patent claims subordinate to patent claims 1 and 15 on the one hand and to the following explanation of preferred embodiments of the invention by way of example, illustrated by the drawing on the other hand. In connection with the explanation of the preferred embodiments of the invention by the aid of the drawing, generally preferred embodiments and further developments of the teaching will be explained. In the drawing is a schematic view illustrating a scenario of a static delivery plan according to prior art, is a schematic view illustrating a system scenario of a fleet management system in accordance with an embodiment of the invention, is a flow diagram illustrating the functional architecture of a fleet management system in accordance with an embodiment of the invention, is a schematic view illustrating the process of time slot generation and time slack introduction in accordance with an embodiment of the invention, is a schematic view illustrating an algorithm for customer reassignment in accordance with an embodiment of the invention, is a diagram showing simulation results for a 10% varying in customer behavior, and is a diagram showing simulation results for a 20% varying in customer behavior.

Fig. 1 schematically illustrates a static delivery plan according to prior art together with various factors that affect the delivery plan and impede its trouble-free execution. As can be obtained from Fig. 1 , a route of the delivery plan (indicated by solid line arrows) starts at a depot and includes several static customers (indicated by solid circles), e.g. as illustrated at Ί '. As illustrated at '2', a customer was missed resulting in an additional idle time. Furthermore, in order to serve the missed customer an additional route is specified that attempts a re-delivery for the missed customer, possibly with serving new customer requests along the new route, as illustrated at '4'. As an additional drawback, illustrated at '3', customers (indicated by open circles) may experience delays in their scheduled deliveries due to external factors, for instance traffic congestion along the route. As a result of these factors that may affect static delivery plans customers may be inconvenienced and the logistic companies may incur additional costs. To resolve these problems, embodiments of the present invention are related to a method to split the delivery planning horizon into time slots based on number of planned deliveries and geographic proximity of customers. Depending on the behaviour of customers, traffic and weather conditions or any factor contributing to dynamicity, each time slot is associated with an appropriate inconsistency factor. The slot inconsistency is then utilized to allot variable slack times to each slot, and the customers are scheduled based on their inconsistency for encountering multidimensional dynamicity. That is, the customers' inconsistency is utilized as a criterion for computing the sequence of deliveries. The method is tailored to handle any spatio-temporal dynamicity which affects the planning.

Fig. 2 schematically illustrates a system scenario of a fleet management system in accordance with an embodiment of the invention. The system, which handles dynamic spatial and temporal changes to meet user expected QoE without overtaxing the operational costs, can be either employed in connection with an existing solution being already in use by the logistic companies or can be treated as a standalone solution. In any case, the system, which may be arranged in a fleet management center 1 , comprises communication means for enabling communication, via a communication network 2, with the vehicles of the fleet on the road. The road network together with the location of the vehicles of the fleet is schematically indicated in the right part of Fig. 2. Communication between the fleet management system and the vehicles of the fleet is bidirectional: In particular, the vehicles of the fleet transmit their log information together with information on any spatio-temporal variations experienced by the vehicles, either continuously or at certain intervals to the fleet management system which, in turn, transmits instructions regarding new or amended routes to the individual vehicles.

As also depicted in Fig. 2, the fleet management system a computing unit 3, which is configured to analyze drive/log information received from the vehicles of the fleet, to analyze information on spatio-temporal variations received from the vehicles of the fleet, and to calculate, as the case may be, new optimized routes, as will be described in more detail in connection with Fig. 3. Although not explicitly shown in Fig. 2, the fleet management system may also comprise interfaces that enable interactions with customers, for instance enabling customers to specify their preferred time windows for their deliveries.

Fig. 3 is a flow diagram illustrating the functional architecture of a fleet management system in accordance with an embodiment of the invention. The embodiment is underlying the following problem statement:

Given a base VRP solution with an associated operational cost Oc and an associated degree of dynamism δ, the objective function is given as: min {Oc {5)} (Eq. 1 ) s.t. QoE of the existing customers is not affected and where Oc (δ) is the operational cost induced due to δ.

First of all, as illustrated at 301 , based on the fleet size and customer locations, an initial delivery plan - base plan -, which serves the demand for deliveries and takes into account existing constraints, is generated after introducing a set of routes. To this end, any of the VRP techniques in the literature may be used (see, for instance, Laporte, Gilbert: "The vehicle routing problem: An overview of exact and approximate algorithms", in European Journal of Operational Research 59, no. 3 (1992): 345-358; or Baker, Barrie M., and M. A. Ayechew: "A genetic algorithm for the vehicle routing problem", in Computers & Operations Research 30, no. 5 (2003): 787-800). The base plan is used as an input to a planning tool, generally illustrated at 302, which aims to address any spatio- temporal variations, as will be described in more detail hereinafter. Basically, at this stage log analysis will be performed in order to optimize the initial delivery plan, thereby taking into account certain optimization goals that can be specified by the respective logistic company. Learning mechanism based on fleet logs to optimize the base plan

The output of the base VRP algorithm, i.e. the initial delivery plan generated as mentioned above by means of a standard VRP algorithm, may be optimal for static customer sets. However, it is very likely that this solution becomes sub-optimal if dynamicity is added to the route each vehicle takes. According to embodiments of the invention, it is envisioned that with the fleet data available, a better cluster of customers can be created. This will help to assign customers in a way that route inconsistencies can be handled in a better way. In a first part of the learning mechanism phase, log-analysis on all customer profiles may be utilized to provide an indication of each customer's reliability. The degree of reliability of a customer may be determined depending on his/her previous record of missed deliveries. A parameter 0 \s used to model the reliability of each customer /based on log-analysis: Θ/ G [0, 1 ], where θ_/ = 0 if the customer is totally unreliable and ^ = 1 if the customer has a record of no missed deliveries. Based on the proximity of Θ_/ ο 1 (highly reliable customer), a customer y^' can also be classified as completely static or quasi-static. The exact classification of customers can be derived from the fleet data analysis, which will help the formation of stable slots. The ratio of completely static to quasi static customers will be used in the dynamic reassignment of the customers, as will be described in more detail further below.

Given the customers i_p = {1 , 2, N}, within a time slot p, the slot inconsistency is proportional to φ_ρ, where:

<PP =∑j≡i_p j (Eq. 2)

Another part of the learning mechanism phase incorporates historical data regarding the total travel time spent, A, for each served delivery to estimate the required time and its associated variation for serving future deliveries with similar characteristics. For such undertaking, log-analysis on the previous records of total time spent for accommodating a number of deliveries within a geographical area and for different traffic conditions is performed. This helps to understand deviations of the actual route time from the planned route time. In turn, this will assist in the inconsistency determination.

In addition, log-analysis for the estimation of traffic conditions, B, in each geographic area during different day times is performed via historical data which contains logs of the position and the speed of vehicles over a significant time period.

Finally, the reactions of vehicle drivers are recorded, and the impact of actions taken by the drivers for any dynamicity which occurs during the execution of the plan on the base operational cost is estimated. For instance, drivers might re- attempt a delivery via inflicting additional costs on another delivery.

Given the log-analysis results, each time slot that includes a number of customers has an associated inconsistency F(A, B, C), where:

A is the statutory time (i.e. the actual route time for a given geographic region), estimated from log analysis, for serving a number of deliveries within the examined time slot,

B is the expected traffic, based on log-analysis, at the examined geographic area during the time period contained in the time slot, and

C is the reaction of drivers on each occurred dynamicity during the execution of the base operational plan. Measuring Customer Quality of Experience (QoE)

Quality of Experience (QoE) is the process of understanding the actual performance of fleet services, as delivered to the customer.

According to an embodiment of the invention, QoE is quantified based on three parameters. The first parameter is the percentage of deliveries executed in the scheduled slot which can be obtained through fleet log analysis and is co-related to factor 'A' (as described above) obtained during log analysis. The second parameter is correlated to the number of customers, Nc, who received their deliveries within the pre-arranged delivery time slot without receiving any notifications about postponing deliveries. The third parameter is the customer retention index, denoted Ci. This index can be obtained from analyzing the logs and determine the percentage of returning customers.

Hence, QoE can be expressed in terms of a function which are governed by Α ' ', Wc'and 'Ci', i.e. f(A, Nc, Ci).

Grouping of customers in time slots and assigning time slacks

Due to working hours constraints, it is assumed that the fleet operator functions during a specific time called operational time 0_T, which starts at t_s and ends at t_e. The actual working time within which fleet vehicles are on road is denoted as Route Time R$ , for each route K, starts at t_s and ends at t_e and is a portion of the actual working time 0_T. Hence, the available buffer time can be used as slack time ASVand is given by:

(Eq. 3)

The customers are served during the total available route time R_T which is split into time slots/periods T_p. These time slots can be of either uniform duration, i.e. customer oriented, or could be adjusted by taking into account the number of customers being served in a geographic region (operation oriented). In order to accommodate the dynamicity of the customers (resulting in particular from missed deliveries and spatio-temporal demand variations), each time slot (corresponding to a sub-route i) is associated with a time slack S . The maximal slack time associated with sub-route i within route /f is given by:

MS_T ^l = ^≠ (Eq. 4) Static time slacks are sub-optimal, because each time slot has different dynamics based on the customer behavior and other factors contributing to dynamicity. A static slack time for each time slot may result in wasted resources, hence, it is favorable to optimize the slack time. According to an embodiment of the invention this kind of improvement is achieved by considering a slack influence factor a associated with each sub-route i.

The optimal Slack time is given by:

OS = CLi · MS (Eq. 5) and 0 < a≤ 1. This slack influence factor a associated with each sub-route i is governed by the slot inconsistency factor SIF_t and is expressed as

SIFi = W x 0_p ) ) + F(A, B, C) (Eq. 6) where:

Nc is the number of customers in sub-route i,

ρ(ψ_ρ) is the degree of dynamicity associated with each customer based on log- analysis, and

F(A, B, C) is the slot inconsistency factor, as introduced above, based on the estimated time spent for each delivery from log-analysis A, estimated traffic conditions B, and drivers' reactions on any inflicted dynamicity C.

Hence,

(Eq. 7) A possible realization of time slot generation and time slack introduction in accordance with an embodiment of the invention is illustrated in Fig. 4.

Dynamic Reassignment of customers and re-optimizing the optimal slack times During the execution phase, the time slotted solution obtained from the base VRP algorithm is monitored continuously for any missed deliveries or spatio-temporal demand changes. If a vehicle V_k encounters a customer C_£ who has some dynamic demands the re-assignment algorithm allots this customer into a different feasible timeslot. During this phase, generally illustrated at 303, instead of executing VRP solution in a classical way (each route starts and ends at the depot), the method utilizes a modified Open VRP approach. Here, it is noted that Open VRP denotes a VRP technique, in which in which vehicles start at the depot but are not required to return to the depot (as described, for instance, in Tarantilis, Christos D., George loannou, Chris T. Kiranoudis, and Gregory P. Prastacos: "Solving the open vehicle routing problem via a single parameter metaheuristic algorithm", in Journal of the Operational Research Society 56, no. 5 (2005): 588-596.). In the described embodiment such Open VRP is modified, such that instead of starting at the depot, the planning can start at any random customer where the vehicle currently is. In other words the vehicles start at a random customer and end at a random customer.

In this approach, the modified open VRP is executed recursively for all the available operational-oriented time slots Tp such that:

Si. Ej E Q where, S_tJ Ej are the starting and end node within the time slot and Q is the set of customers with static behavior index. In other words, the modified open VRP method operates in each time slot by starting and ending the route at a random node based on the customer behavior index.

During the execution, the algorithm finds an appropriate slot for an attempted redelivery for customer C_£ based on the operational cost minimization, as indicated at 304. After defining the timing of the rescheduled delivery, a notification for the delivery waiting time is given to the customer. The waiting time is equal to the customer-oriented time slot (i.e., 1 hour). During the re-assignment phase, the algorithm tries to stay strongly committed with static customers so that the time slot time T_p for them is not affected, in order to accommodate the spatio-temporal demand of customer C_£. A flow chart of the algorithm is shown in Fig. 5.

During the reassignment, each slot is checked for the ratio of completely static to quasi static customers. The re-assignment algorithm can either retain the quasi- static customer in the same slot or move the customer to another slot if better OPEX is achieved, as shown in Fig. 5, where an affected/missed delivery to customer C3 is re-assigned from the initial time slot T+1 ' to a new time slot '1+2'. The boxes surrounding customers C1 -C5 (initial plan) and C2-C5 (online optimized plan), respectively, indicate the remaining total time-period of the whole plan. The dotted line boxes indicate the customers who have already been served. This approach helps to optimize further the OPEX of fleet operators, without penalizing the reliable customers.

During the execution of the plan, with every successful delivery the associated slot inconsistency factor SIF_t is decremented and a new OS_? is associated. By dynamically updating the slack time (and re-adjusting the time-slacks during runtime, in particular based on spatio-temporal data), the Route time is reduced considerably, thereby allowing the fleet operators to schedule more deliveries within the operational time.

It is noted that re-scheduling of customers is not restricted to change events. The customers can also be re-scheduled if the proposed system during execution evaluates a new plan is beneficial for the overall OPEX. As mentioned before, any change in the plan considers the QoE of the customers.

In order to minimize the effect of dynamism on the OPEX, once the overall slot inconsistency is determined, embodiments of the proposed invention attempt to schedule customers with higher inconsistency at the beginning of the slot times. To perform this action, the inputs of the open VRP algorithm are modified. Customers are associated with a weight factor, W(O), and the higher the inconsistency, the lower the weight factor. The open VRP algorithm computes the sequence of deliveries based not only on transport-related parameters, such as the distance of customers and the travel times, but also on the associated weight factor of each customer.

The above approach can also be applied to supply chains logistics, with single source and destination. A key effect in supply chain logistics is a pertinent domino effect which is spread across the various entities. The method proposed in accordance with the invention can be used to identify such effect and adjust the times to minimize the effect.

An embodiment of the present invention was tested on Solomon's Test Instance. Solomon's are best-known solutions for 25 - 200 customer instances. Preliminary results (illustrated in Fig. 6 for 10 % and in Fig. 7 for 20 % missed deliveries) show that:

• The travelled distance is reduced by 20% compared to rescheduling missed customers to the next day

· Re-planning the missed customers with other set of customers leads to wasted capacity and additional routes.

Therefore, the proposed invention improves the OPEX compared to the current rescheduling systems in terms of costs incurred and the required resources.

Many modifications and other embodiments of the invention set forth herein will come to mind the one skilled in the art to which the invention pertains having the benefit of the teachings presented in the foregoing description and the associated drawings. Therefore, it is to be understood that the invention is not to be limited to the specific embodiments disclosed and that modifications and other embodiments are intended to be included within the scope of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.

Claims

C l a i m s

1. Method of dynamic fleet routing,

wherein said fleet includes a number of vehicles that perform delivery services, and

wherein an initial delivery plan is generated based on a customer demand for deliveries, wherein said initial delivery plan specifies routes for said vehicles and schedules customers of the delivery services within the routes,

c h a r a c t e r i z e d i n the steps of

splitting the planning period of said initial delivery plan into a number of time slots,

associating each of said time slots with a slot inconsistency factor depending on factors that contribute to dynamicity within the respective time slot, and

2. Method according to claim 1 , wherein said initial delivery plan is generated by applying a VRP (Vehicle Routing Problem) algorithm.

3. Method according to claim 1 or 2, wherein said time slots are generated as customer-oriented time slots by splitting the planning period of said initial delivery plan into a number of time slots of equal lengths.

4. Method according to any of claims 1 to 3, wherein said time slots are generated as operational-oriented time slots by splitting the planning period of said initial delivery plan into a number of time slots having lengths proportionally to the number of customers in close proximity.

5. Method according to any of claims 1 to 4, wherein customers are offered to select, from the number of generated time slots, time slots they prefer for their deliveries.

6. Method according to any of claims 1 to 5, wherein said factors that contribute to dynamicity within a time slot and that are taken into account for determining said inconsistency factor include the customers' reliability.

7. Method according to any of claims 1 to 6, wherein a customer's reliability is derived from log-analysis of recorded customer profiles.

8. Method according to any of claims 1 to 7, wherein a customer is assigned a quasi-static behavior index in case his reliability exceeds a predefined configurable threshold.

9. Method according to any of claims 1 to 8, wherein customers with higher inconsistency are scheduled at the beginning of said time slots.

10. Method according to any of claims 1 to 9, wherein said factors that contribute to dynamicity within a time slot and that are taken into account for determining said inconsistency factor include at least one of the factors:

the estimated time for serving a number of deliveries within said time slot, the expected traffic at the examined geographic area during the time period contained in said time slot, and

the reaction of drivers of said vehicles on occurred dynamicity during the execution of initial delivery plan.

1 1. Method according to any of claims 1 to 10, wherein the lengths of said individual slack times allotted to said time slots are designed to be proportional to said inconsistency factor associated with the respective time slot.

12. Method according to any of claims 1 to 1 1 , wherein the quality of experience for the customers is determined as a function of the percentage of deliveries executed in the scheduled time slot, the number of customers who received their deliveries within the pre-arranged delivery time slot, and a customer retention index.

13. Method according to any of claims 1 to 12, wherein a continuous monitoring is performed for any missed deliveries and/or spatio-temporal demand changes.

14. Method according to any of claims 4 to 13, wherein an open VRP is executed recursively for all existing operational-oriented time slots of a route such that the first customer and the last customer within each of said operational- oriented time slots are customers having assigned a static or quasi-static behavior index.

15. Fleet management system, in particular for executing a method according to any of claims 1 to 14,

c h a r a c t e r i z e d i n that the system comprises computation means that are configured

16. System according to claim 15, further comprising communication means for sending instructions to and receiving log information from the vehicles of the fleet.

17. System according to claim 15 or 16, further comprising a log analyzer for analyzing log information on customer's profiles, in particular with customers' missed delivery records, on the total travel times spent for accommodating a number of deliveries within a geographical area and for different traffic conditions, on traffic conditions in different geographic area during different day times, and/or on the behavior of vehicle drivers in reaction to dynannicity.

18. System according to any of claims 15 to 17, further comprising interfaces for enabling interactions with customers.