Optimizing Hydrogen Refueling Station Locations and Tour Planning: Heavy Weight Road Transport in the Northern Netherlands

(1)

Optimizing Hydrogen Refueling Station Locations

and Tour Planning:

Heavy Weight Road Transport in the Northern

Netherlands

January 24, 2021 Joris Hoekman - S2917297

The University of Groningen

Abstract: Within the road transport sector, heavy weight vehicles (HGV) are

responsible for a significant share of greenhouse gasses worldwide. Hydrogen fuel cell technology is assumed to play a key role in decarbonizing the HGV sector. However, the lack of a refueling infrastructure is one of the major constraints to the widespread adoption of hydrogen fuel cell vehicles (HFCVs) in the HGV context. Hence, the development of such infrastructure is required, in which both location decisions (for refueling stations) and vehicle routing decisions should be considered. (Vehicle routing decisions relate to the process of planning vehicle routes conforming to transport orders.) Both of these aspects are addressed in this study. In reality many deliveries are made on round-trip basis. Thus it is important to combine both location decisions and tour planning in one model in order to obtain optimal results. Therefore, this paper aims to validate a model in which tour planning and locations decision are simultaneously optimized, through its application on a real-life case in the northern Netherlands. The results of this study show that in the analyzed case the transition to a fleet of HFCVs is operationally feasible. On the other hand, the financial feasibility remains equivocal. Lastly, the study shows that the model can be used as a decision support tool on strategic route planning for businesses. This paper is, to the knowledge of the author, the first to research the location routing problem in a real-life European case study focusing on heavy weight transport using HFCVs.

Keywords: Facility Location Problem, Vehicle Routing Problem,

(2)

J.W.Hoekman

Master’s thesis - MSc Technology and Operations Management Supervisors: dr. E. Ursavas & dr. X. Zhu

(3)

J.W.Hoekman CONTENTS

Optimizing Hydrogen Refueling Station Locations and

Tour Planning:

Heavy Weight Road Transport in the Northern

Netherlands

Joris Hoekman

1 Introduction

(6)

implies that the explosive power per volume is rather low for hydrogen and thus safer compared to other fuels such as diesel (Adamson & Pearson, 2004) [1].

Major obstacles still stand in the way of widespread HFCV adoption. Firstly, the vehicle costs of HFCVs are still very high. According to Moultak et al. (2017) [3] the current capital costs of a hydrogen fuel cell trailer-truck in Europe are about €232.000. These capital costs are 21% of the total cost of ownership (TCO) of HFCVs in contrast to 11% of the TCO of conventional vehicles (Deloitte, 2020) [4]. Secondly, the high costs of hydrogen obstruct the adoption of HFCVs (about €12-15/kg today (Zhang et al., 2020) [5]). And lastly, the lack of the hydrogen refueling infrastructure prevents the widespread adoption of HFCVs.

Constructing the hydrogen refueling infrastructure requires huge investments. HRS opening costs

can range from€1 - €4.5 million (Grant Agreement HEAVENN, 2019) [6]. One of the key factors

determining these HRS costs is the station capacity. For example, for HRSs with a capacity of 160 kg/day the total capital costs are assumed to be€2.19 million and for stations with large capacities of 1500 kg/day€4.18 million (Melaina, 2003) [7]. Due to budget constraints, only a limited number of hydrogen refueling stations (HRS) can be built. Moreover, in the beginning, low utilization of the infrastructure will be unavoidable. For this reason, the strategic placement of HRSs is essential. To be competitive with conventional vehicles, routing decisions of HFCVs are important as well. Whereas conventional trucks have many opportunities to refuel along the way, HFCVs do not have this convenience due to limited HRS availability. Therefore, proper calibrated tour planning is re-quired in order to ensure that desired destinations can be reached by HFCVs just like conventional

vehicles. Moreover, due to the high vehicle acquisition costs of HFCVs (about 36€/100 km for

HFCVs compared to 10 €/100 km for conventional HGVs (Deloitte, 2020) [4]), high fleet utiliza-tion is essential to cut down operautiliza-tional costs and be competitive with convenutiliza-tional vehicles. To ensure high fleet utilization, cost or time optimized tour planning is key.

This study builds on prior literature from two streams of research: those considering the facility location problem and those considering the vehicle routing problem, both of which are well estab-lished fields of study. Within the area of facility location problems, the flow refueling location model (FRLM) (Kuby & Lim, 2005) [8] is focused on locating refueling stations. This model identifies the number of refueling stations that maximize the traffic flow between origin-destination (O-D) pairs, whilst considering the limited driving range of vehicles. For the vehicle routing problem (VRP), the goal is to identify the optimal route for a fleet of vehicles visiting a certain set of locations (the classical VRP). An important variation of this problem is the Green VRP (Erdogan & Miller-hooks, 2012) [9]. The objective remains the same as for the classical VRP; however, additional challenges resulting from operating a fleet of alternative fuel vehicles (e.g. HFCVs) are considered.

(7)

independently regularly leads to sub-optimal results. Moreover, in location planning it is often assumed that deliveries are made on a direct-trip basis, while in reality many deliveries are made on a round-trip basis. Hence, for those cases it is required to combine the location of a depot or station and the tour planning in one model (Schwardt & Fischer, 2009) [11]. A model that combines these location and routing decisions is proposed by Kamer (2017) [12].

Even though extensive research efforts have been undertaken in both directions, little attention has been paid to alternative fueled truck delivery contexts and application of real-life HGV cases. Two of the few studies that consider specific HGV contexts are Kluschke et al. (2020) [13] and Baek et al. (2020) [14]. Kluschke et al. (2020) consider solely facility location problems, whereas Baek et al. (2020) base their research on vehicle routing literature.

Despite some efforts (Kluschke et al., 2020; Rose & Neumann, 2020) [13][15], the application of these models to European data remains very limited. To stimulate adoption of HFCVs in Europe, it is particularly important that such European infrastructure maps are developed. Thereupon, business cases and road-maps can be built. Therefore, the aim of this study is to apply the model proposed by Kamer (2017) [12] on a European case, thereby furthering the adoption of HFCVs in Europe. The focus area of this research is located in the northern Netherlands; the so-called “HEAVENN” region.

The research question this study aims to answer is as follows:

What are optimal hydrogen refueling station locations and optimal routes for heavy goods vehicles in the northern Netherlands?

Data from the northern Netherlands concerning HGV transport, HFCV specifications and infras-tructural aspects are gathered and constructed. With the use of a mixed integer program, the model is solved.

The main contributions of this paper are the collection of data and application of the model on a real-life case. These contributions will help further the adoption of HFCVs in Europe. This paper differs from others concerning the transport segment of HGVs (in contrast to the more commonly researched passenger cars) and the analyzed region (northern Netherlands). Hence, the gap in literature is filled by researching the location routing problem in a real-life European case study focusing on heavy weight transport using HFCVs.

(8)

J.W.Hoekman 2 Theoretical Background

2 Theoretical Background

The aim of this study is to provide an optimal HRS infrastructure and routing planning for vehicles in the HGV sector in the northern Netherlands. This research combines two streams of literature: the Vehicle Routing Problem (VRP) and the Facility Location Problem (FLP). First the vehicle routing literature and facility location literature are discussed. The combination of the two literature streams is also known as the Location Routing Problem (LRP), which will be discussed as well. Lastly, the contributions of this paper are addressed.

2.1 Vehicle Routing Problem

The classical vehicle routing problems poses the question what the optimal set of routes will be for a fleet of vehicles to deliver from a depot location to a set of customers. As the classical VRP is a basic model, it has been extended in several ways. Two seminal extensions are the considerations of capacities and time windows. The capacitated VRP (CVRP) (Dantzig & Ramser, 1958) [16] adds to the classical VRP by considering maximum vehicle capacities. Conrad (2011) [17] extended the CVRP with time windows to depots and customers. These seminal concepts have been deepened over the years and have been adapted to alternative fueled vehicle context.

One of the first studies to include alternative fuel vehicles to the VRP is Erdogan & Miller-Hooks (2012) [9]. Their Green VRP deals with the problem of minimizing the distance travelled while considering the limited range of alternative fuel vehicles and unbalanced distribution of alternative fueling stations. Zhang et al. (2020) [18] also consider alternative fuel vehicles (both HFCV & EV). However, their objective is to minimize total carbon emissions rather than costs. Later, Schneider, Stenger, and Goeke (2014) [19] adapted the Green VRP by considering EVs only and by introducing time window constraints (E-VRPTW). As EV recharging times can be very long, the time window constraint is a very appropriate addition to the model. A subsequent extension on the work by Schneider et al. (2014) [19] introduces a heterogeneous fleet to the concept (Hiermann et al., 2016) [20]. Vehicle types may vary in load capacity, battery size and acquisition costs. Therefore, the consideration of a fleet with different vehicle types may produce more realistic routing results.

2.2 Facility Location Problem

(9)

J.W.Hoekman 2.2 Facility Location Problem

as a more static element, whereas the FRLM assumes that demand is flow-based. As the refuel-ing of HGV transport is best characterized as a demand flow, this study follows the FRLM approach. The flow-capturing location allocation model (FCLM) (Berman et al., 1992; Hodgson, 1990) [22][23] is one of the first models that was based on the idea of demand flow. The FCLM model locates facilities to serve demand flows to the highest extent. The model assumes facilities not to be the destination of a trip (e.g. ATMs, billboards, refueling stations). Rather, demand is considered as a flow that traverses these facilities (e.g. traffic flow). Unlike stationary demand, demand flows have an origin and destination (O-D). The flow is assumed to be “captured” if at least one facility is sited on the shortest path between the O-D nodes of the flow.

Kuby and Lim (2005) [8] extended the FCLM to the flow refueling location model (FRLM). They designed a model to find optimal locations of refueling facilities for alternative-fuel vehicles consid-ering the limited range of these vehicles. In contrast to the FCLM, vehicles can refuel more than once at different facilities if necessary. Here, demand flows are only considered to be served if there are sufficient stations placed along the pathway. The FRLM can be used in two ways. It can either maximize the vehicle trips covered when locating a fixed number of stations in a network (maximum covering), or minimize the number of facilities needed to cover a given demand share (set covering) (Jochem et al., 2016; Kluschke et al., 2020) [24][13]. To solve the model, pre-determination of all valid station combinations on each path is required.

The pre-processing stage of the FRLM is time consuming, which makes it hard to solve larger networks. Accordingly, Lim and Kuby (2010) [25] introduced heuristic algorithms to deal with such sizeable problems. Later, Capar & Kuby (2012) [26] (NC-PC) and MirHassani & Ebrazi (2013) [27] developed even faster solution methods for the FRLM. Both studies used mixed integer linear programming (MILP) models that do not require pre-generation of feasible station combinations. In addition, the two models can solve both the set covering and maximum covering problem. In contrast to the node-cover path-cover (NC-PC) model of Capar & Kuby (2012) [26], the arc-cover path-cover (AC-PC) model (Capar et al., 2013) [28] focuses on covering the arcs of the path rather than the nodes. The unique feature of this model is the reformulation of the FRLM by introducing a new set of candidate stations for each arc. Like previous studies, this model does not require any pre-generation of feasible station combinations. Differently, the AC-PC has less constraints and variables like previous studies. Hence, the model is much more compact, and it is computationally faster.

(10)

J.W.Hoekman 2.3 Location-Routing Problem

2020) [13][15]. The first one introduces node capacity restrictions as an extension, the second study researches the interplay between the power system and station networks. Both studies apply their model to real-life case studies and networks in Germany.

2.3 Location-Routing Problem

Facility location problems and vehicle routing problems have been addressed separately for a long time. However, there is growing attention for the combination of both problems: the location-routing problem (LRP). The classical LRP exists of opening a set of depots, assigning customers to these depot and determining the routes of the vehicle. The objective of the classical LRP is to minimize the total costs, which consist of the costs of opening depots, fixed costs (or acquisition costs) of the vehicles and the total costs of the routes.

As mentioned before, the first authors to identify the potential benefits of the location-routing problem combination showed that solving location and routing problems independently may lead to sub-optimal results (Salhi & Rand, 1989) [10]. More recently, Darvish & Coelho (2018) [33] studied a production-distribution system considering location, production, inventory and distribu-tion decisions. Similar to Salhi & Rand (1989) [10], they found that significant enhancements are realizable when simultaneously optimizing for locations and routing decisions.

Similar to the VRP and FRLM, the LRP has uncapacitated versions (Hashemi Doulabi & Seifi, 2013; Tuzun & Burke, 1999) [34] [35] and capacitated variants considering capacities on locations (Baldacci et al., 2011; Belenguer et al., 2011) [36] [37]. An LRP model which includes environ-mental impacts is introduced by Wang et al. (2020) [38]. This model considers green logistics and life-cycles of eco-packages. Zheng et al. (2019) [39] study the integrated optimization of vehicle routing, location decisions and inventory in a supply chain network design.

(11)

J.W.Hoekman 2.4 Contributions Table 1: Literature Comparison

Research papers Research Aspects VRP FLP LRP (VRP & FLP) HFCV Practical Case Study Heavy Duty Transport

Schiffer & Walther (2017) _X

Kamer (2017) _X _X

Hiermann et al.(2016) _X

Rose & Neumann (2020) _X _X _X _X

Kluschke et al.(2020) _X _X _X _X

Baek et al. (2020) _X _X

Zhang et al.(2017) _X _X

Sterzik et al.(2011) _X _X

Liu & Wang (2017) _X

Zhang et al.(2020) _X _X

Yang & Sun (2015) _X

Kang & Recker (2015) _X _X _X

Zheng et al.(2019) _X _X

This paper _X _X _X _X

2.4 Contributions

Like most research on the LRP, Kamer (2017) [12] explores solution approaches to the developed model. Real-life data is not included in Kamer’s model analysis. In contrast, this study does in-clude real-life case data; the Hydrogen Location Routing Problem (H-LRP) model (Kamer, 2017) [12] is used and applied on a case in the northern Netherlands. The paper of Kamer contributes to the academic literature by extending the E-FSMFTW with locations decisions, options for partially recharging and the consideration of a heterogeneous fleet of HFCVs. Using a mixed integer program-ming model and a column generation algorithm the problem is defined and solved. The research by Kamer showed the importance of considering both locations decisions and partial recharging in the specific problem setting.

(12)

J.W.Hoekman 3 Methodology

3 Methodology

In this section the model formulation and data collection procedure are introduced. As this study attempts to optimize infrastructures and tour planning, quantitative model-based research is deemed the most appropriate method. As mentioned before, the mixed integer linear programming model (MILP) introduced by Kamer (2017) [12] is used in this research. No significant modifications have been made to his model.

3.1 Model Formulation

3.1.1 Model Assumptions

The assumptions underlying this model are:

1. A vehicle cannot travel from one refueling station to another refueling station; 2. Vehicles are fully refuelled when setting out from the depot;

3. Only highway networks are considered to be part of the network;

4. Only HGVs are considered, motor cycles or passenger cars are excluded; 5. HGVs travel on average 70 kilometres per hour (43.5 miles per hour).

3.1.2 Parameters & Variables

The variables and parameters used in the model are presented below. The model notation and mathematical representation by Kamer (2017) [12] are derived from Hiermann et al. (2016) [20].

Parameters

C set of all customers.

F set of all refueling stations.

F_j0 duplicates of refueling station j ∈ F . F0 set of all duplicated refueling stations. b, (e) depot node for the start(end) of a route.

B set of all depot nodes.

V set of all vertices, given by V = {b} ∪ {e} ∪ C ∪ F0. VB start depot nodes, given by VB = {b} ∪ C ∪ F0.

VE end depot nodes, given by VE = {e} ∪ C ∪ F0.

VP set of nodes excluding the depot, given byVP = C ∪ F0.

E the edge set on the vertices in V .

K set of vehicle types.

Yk _{fuel capacity of vehicle type k.}

(13)

J.W.Hoekman 3.1 Model Formulation

rk_(·) _{fuel consumption as a function of the current load of vehicle type k,} given by rk_{(q) = (1 + α}k_q)ˆ_rk_.

ˆ

rk _{fuel consumption of an empty vehicle of type k.}

αk load-dependence coefficient of vehicle type k.

Qk load capacity of vehicle type k.

fk _{acquisition cost of vehicle type k.} [`i, ui] time window at node i.

si service time at node i.

dij distance from node i to j.

tij travel time from node i to j.

Ui cost of opening a station at location i ∈ F .

pj demand at customer node j ∈ C.

Variables xk

ij binary decision variable that indicates a vehicle of type k travels from node i to node j. zi continuous decision variable giving the fuel recharged at when visiting i ∈ F0.

wi binary decision variable that indicates the station at location i ∈ F is opened. τk

i start of service of a vehicle of type k in node i. qk

i current load of a vehicle of type k in node i.

yk

i current fuel level of a vehicle of type k in node i.

3.1.3 Mathematical Model

The mixed-integer linear program model is given by:

(14)

J.W.Hoekman 3.1 Model Formulation 0 ≤ yk_j ≤ yk b − r k_(qk j + pj)dbj+ M (1 − xkbj) ∀j ∈ VE, ∀k ∈ K (3.10) 0 ≤ yk_j ≤ yk_i − rk(q_ik)dij + M (1 − xkij) ∀i ∈ C, ∀j ∈ VE, i 6= j, ∀k ∈ K (3.11) 0 ≤ yk_j ≤ yk i + zi− rk(qik)dij + M (1 − xkij) ∀i ∈ F 0 , ∀j ∈ VE, i 6= j, ∀k ∈ K (3.12) y_bk= Yk ∀k ∈ K, ∀b ∈ B (3.13) 0 ≤ zi ≤ Yk− yki ∀i ∈ F 0_{, ∀k ∈ K} _(3.14) xk_ij ∈ {0, 1} ∀i ∈ VB, ∀j ∈ VE, i 6= j, ∀k ∈ K (3.15)

The Big-M method, where M is defined as:

M = max k∈K ( Yk+ rk X i∈C pi ! ¯ d ) (3.16) ¯ d = max i∈VB,j∈VE {dij} (3.17)

The objective function is given by equation (1), which aims to minimize the costs of opening sta-tions, the vehicle acquisition costs, and the total distance travelled. The total distance travelled can be used as a direct measure for fuel costs by taking the product of the distance travelled the fuel consumption rate and the fuel price per fuel unit. Both the fuel consumption rate and fuel price are assumed to be fixed. Hence, the total distance travelled can be used to minimize costs and to make the model find optimal routes. Constraint (2) guarantees that each customer is visited exactly once by a vehicle. (3) ensures that a refueling station can only be visited if the station is open. Constraint (4) guarantees that vehicles that visit customers will as well leave the customer. The time windows of the model are addressed with constraints (5-7). Here (5) indicates that the start of a service needs to be within the time window of a customer [li, ui]. (6) assures the chronology of events: a service at a node cannot start before the service time si and travel time tij from the previous customer have elapsed. Constraint (7) guarantees that if the previous node was a refueling station the refueling time of a vehicle gk_z

i is used rather than the service time.

Constraint (8) guarantees the demand of customers is fulfilled and (9) ensures that the loads of vehicles are non-negative and cannot exceed the maximum load capacity of the vehicle.

The constraints (10–14) represent the fuel levels of the vehicles. (10) ensures that the fuel lev-els are consumed when travelling to a node from the depot and (11) does the same for travelling to a node from a customer. Constraint (12) constitutes the fuel levels after a refueling station has been visited. Upon departure of the refueling station, the fuel level of the vehicle equals yk

i + zi. If the next node to be visited is j, the fuel level depletes with rk_(qk

i)dij.

(15)

J.W.Hoekman 3.2 Data Collection

maximum capacity when a vehicle sets out from the depot. (14) ensure that the fuel-cell cannot exceed its maximum fuel capacity. The last constraint (15) assures that the decision variable xk

ij is a binary variable. Furthermore, the model makes use of the “Big-M method”. This method helps solving the linear programming problem. M is defined by equation (17).

The mathematical model will be solved using the Python programming language and the Gurobi solver.

3.1.4 Preprocessing of the Network

In order to make the network suitable for the model, two preprocessing steps are required. For all the edges of the network (edge set E), any edge needs to be removed between two recharging stations. Hence, a vehicle is not able to travel from one recharging station to another. Moreover to eliminate infeasible customer series, remove the edge between customer i and customer j if:

li+ si+ tij > uj or (3.18)

pi+ pj > Qk, ∀k ∈ K or (3.19)

rk(pj)dij > Yk, ∀k ∈ K (3.20)

Customer series can be infeasible in three ways. First, if the sum of the start of the time window at node i, service time at node i and travel time between nodes i & j exceed the end of the time window at node j (3.18) the customer sequence is infeasible. In other words starting, providing service and travelling from node i requires too much time to make it on time at node j. Secondly, if the demand for node i and for node j is larger than the vehicle capacity (3.19) customer sequences are infeasible. This implies a vehicle is not able to transport all demand itself to both nodes. Third, if the fuel required for travelling from node i to j is higher than the fuel capacity of the vehicle (3.19) the customer sequence is infeasible.

3.2 Data Collection

To solve the model, different types of case-specific input data are required:

• _{Information such as nodes of the network and distance between adjacent nodes is required to} describe the highway network (vertices, start depot nodes & end depot nodes).

• _{Data on the traffic flows of HGV vehicles and the service; customer demand, time windows,} service times (Ministerie van Infrastructuur & Milieu, 2020)[43].

• _{Specifications on HFCVs; like vehicle range (fuel capacity), load capacity, acquisition costs} and fuel consumption rates (Hyundai, 2020; Holthausen Clean Technology, 2019; Deloitte, 2020)[44][45][4].

(16)

J.W.Hoekman 3.2 Data Collection

• _{Costs of opening a HRS station at a certain location (Melaina, 2003) [7].}

(17)

J.W.Hoekman 4 Scenario Analysis

4 Scenario Analysis

In this chapter the model presented in chapter 3 is applied to a real-life case: the HEAVENN project. First, the HEAVENN case and the network layout derived from it are introduced. Second, two business case scenarios are introduced to investigate the adoption of HFCV transport. From the perspective of two businesses that are active in heavy weight transport in the northern Netherlands, we evaluate the financial and operational aspects of HFCV adoption.

4.1 HEAVENN Project

The focal case of this study is the HEAVENN project, a large-scale hydrogen infrastructure project in the northern Netherlands. The northern Netherlands is the first to receive a European grant for the development of a functional green hydrogen infrastructure. The project started in January 2020 and is expected to run until the end of 2025 (Hydrogen Valley, 2020) [46].

The grant has been allocated to the northern Netherlands, because this region is well-suited to develop a hydrogen economy. Not only is there a high availability of renewable energy sources, the region is also characterized by adequate storage and transport opportunities, its industrious nature, mobility and built environmental hydrogen applications. All sub-endeavours of the HEAVENN project are focused on sectoral integration. The main goal of this project is to make use of green hydrogen across the entire value chain, while developing replicable business models for wide-scale commercial deployment of hydrogen across the entire regional energy systems (European Commis-sion, 2020) [47].

Among the tasks of the HEAVENN project is the exploration of heavy duty mobility applications in the northern Netherlands (Grant Agreement HEAVENN, 2019) [6]. This includes case-specific infrastructural aspects such as the transport network, refueling stations and HFCVs utilized in the HEAVENN region.

4.1.1 Transport Network

(18)

J.W.Hoekman 4.1 HEAVENN Project

The network is depicted in figure 1.

The path distance between nodes is estimated using the route planner of Google Maps. Here, the start and end of the highway for each node are considered, respectively as start and end points of the pathway. Using Dijkstra’s algorithm (Dijkstra, 1959) [48] the shortest pathway between each node can be calculated (Appendix B.1). As we assume that HGVs travel with an average speed of 70kph (43.5 mph), the travel time of the shortest pathway can be calculated as well.

Similar to Kamer (2017) [12], the fuel load dependency is not considered in this research. From a modelling perspective, this implies that rk is taken into account instead of rk(·).

Figure 1: Highway Network - HEAVENN Region

4.1.2 Hydrogen Refueling Stations

(19)

These two standards of hydrogen compression (350 bar and 700 bar fuel dispenses) have pros and cons. The advantage of the 700 bar storage is that a higher vehicle range can be realized with fewer storage space (on the vehicle) compared to the 350 bar storage. This implies that for 350 bars storage more space is required for hydrogen storage tanks on the vehicle to achieve a similar range.The disadvantage of 700 bar storage is financial; the costs of high pressure storage are much higher than for lower storage pressures. For these reasons, high pressure levels of 700 bars are often used by smaller vehicles, whereas the 350 bars pressure levels are often used by HGVs, as these vehicles do have more on-board space for fuel storage.

Besides different levels of compression, HRS have also differences in refuelling speeds. If several vehicles need to be filled at a HRS, then fast filling stations are most appropriate. If only one or few vehicles need refueling, slow filling stations may be more appropriate. In general, faster refu-elling requires high pressure storage. Therefore station costs for fast filling stations are often higher. The HEAVENN project includes plans for improving the HRS availability. Four HRS locations are considered in these plans: Delfzijl, Pesse, Groningen & Emmen (Grant Agreement HEAVENN, 2019) [6].

In Delfzijl there is an existing PinPoint 350 bar HRS. There will be an upgrade of the existing HRS to add 700 bar capability and increased capacity to refuel cars, LGVs and HGVS. The station currently already provides hydrogen fuel for two public buses, after expansion it will be able to de-liver an additional 100kg of hydrogen per day at 700 bar (Grant Agreement HEAVENN, 2019) [6]. This HRS is supplied by a pipeline from the nearby chlor-alkali facility where hydrogen is produced. In Pesse a GreenPlanet 350bar HRS has been realised. The HRS will be upgraded to 700 bar capability to refuel medium and heavy HFCVs as well. The main supply method for this HRSs is trailer truck delivery (Grant Agreement HEAVENN, 2019) [6].

In Groningen there will be a deployment of an additional 350 bar HRS, which will be able to deliver approximately 600 kg/day. Besides this new HRS, the already existing 350 bar HRS will be increased to 700 bar capability to refuel cars, LGVs, HGVs and garbage trucks. This HRS will have an estimated dispensing capacity of approximately 1,000 kg of hydrogen per day. For example, if trucks with an on-board storage tank of 30kg were to be deployed, the total number of trucks that can be refuelled on daily basis in Groningen is (1600/30) about 53. The main supply method for the HRSs in Groningen will be trailer truck delivery. The delivery method of hydrogen is expected to move from truck delivery to pipeline supply as the volume grows and the existing natural gas infrastructure becomes re-purposed (Grant Agreement HEAVENN, 2019) [6].

(20)

HEAVENN, 2019) [6]. Similar to the stations in Groningen, the HRS will at first be resupplied by hydrogen truck trailers. As hydrogen volumes grows, pipeline delivery may be employed.

In this research these locations are included as potential station locations. However, Emmen is excluded as the HRS here will only be available for public bus transport refueling and not for HGVs. As an experimental setting we include Heerenveen as potential refueling location for HGVs, due to its central position in the transport network. Moreover, in this research it is assumed that all HRSs are fast filling stations.

4.1.3 Hydrogen Fuel-Cell Vehicles

The HEAVENN projects sets the goal to deliver 4 garbage trucks and 6 HGVs (approx. 105 tpy) as hydrogen end-use. However, as the development of hydrogen trucks is a recent and novel area, availability of hydrogen fuel-cell heavy duty trucks is limited. Two different hydrogen truck suppliers and their developed HGV are presented in Table 3: Hyundai and Holthausen (Converted DAF-CF).

Table 3: HFCVs specifications

Manufacturer Hyundai XCIENT [44] Holthausen [45]

Range 400 km 400 km

Power total 350kW ?

H2 Storage 32,09kg @ 350 bar 20 - 40 kg @ 350 bar

Battery 73,2 kWh 140 kWh

(21)

J.W.Hoekman 4.2 Scenario Formulation

4.2 Scenario Formulation

The financial and operational feasibility (business case) of HFCV adoption is explored for two scenarios. The aspects under consideration are given in Table 4. By employing the model presented earlier,the routing and HRS location feasibility are evaluated. Regarding the financial aspects, the vehicle acquisition costs and opening costs of HRSs are considered. Fuel costs and maintenance costs are not included in the model, but will be considered in the business case for each scenario.

Table 4: Business Case Aspects

Operational Aspects Financial Aspects

Routing Vehicle Acquisition Costs

HRS Locations HRS Opening Costs

Fuel Costs Maintenance Costs

4.2.1 Scenario 1 - Supermarket Logistics

In the first scenario, a Dutch supermarket (Jumbo Groep Holding B.V.) is examined. This company owns 687 supermarkets across the Netherlands, 83 of which are located in the northern Netherlands. This study focuses on the northern distribution centre of this particular company, which is located in Beilen. From there, heavy duty trucks depart to resupply the supermarkets in the northern Netherlands. The feasibility of transitioning from fossil fuelled trucks to HFCVs is evaluated from the perspective of the business owner. The actual supermarket locations are used in the analysis, though the demand data is based on assumptions.

In this scenario, only supermarkets are considered which are positioned in the cities (nodes) of the HEAVENN region network (Figure 1). It is assumed that the demand for supermarket replen-ishment depends on the size of the supermarket and the day of the week (weekdays and weekends). On average, these supermarkets are resupplied two times per day by a trailer truck (Ministerie van Infrastructuur & Milieu, 2020) [43]. This amounts to a total of 35 supermarkets needing two-daily replenishment. The supermarkets per city (node) can be found in Table 5. Hence, a total of 70 trailer truck trips are required. In reality, the number of trips required will be higher, as there are more supermarkets located across the northern Netherlands than those based in the network nodes. However, as described earlier in this paper, the model is limited to the inclusion of the highway network. Since regional supermarkets will be replenished primarily by trucks travelling on regional roads, these supermarkets cannot be included in the model. Furthermore, a single truck visits mul-tiple supermarkets to resupply and upon completion of its deliveries it returns to the distribution centre in Beilen.

(22)

J.W.Hoekman 4.2 Scenario Formulation Table 5: Supermarket Resupply

Node Supermarkets per City Deliveries per day Demand ρ

Assen 4 8 11% Groningen 7 14 20% Delfzijl 2 4 6% Drachten 2 4 6% Heerenveen 2 4 6% Leeuwarden 8 16 23% Sneek 2 4 6% Meppel 2 4 6% Hoogeveen 1 2 3% Emmen 5 10 14% Total 35 70 100%

the problem for a single truck with a load devoted to one supermarket. However, a round-trip to multiple nodes without refueling may be problematic. Hence, round-trips with multiple supermar-ket visits need to be checked to ensure all trucks are able to make their trips. The route with the highest number of supermarket visits will be the one in which one truck is required to replenish all supermarkets in the network (e.g. consider the case a chiller lorry needs to deliver frozen products to each supermarket in the network). Using the model presented earlier this instance is tested. Input Data Scenario 1

The specifications of the HFCV are shown in Table 6. These specifications are based on the Hyundai XCIENT fuel cell heavy-duty truck (Hyundai, 2020) [44]. The fuel capacity of this vehicle is claimed to be 32.09 kg of hydrogen. The fuel consumption rate is assumed to be 0.08. HFCVs are assumed to travel 70 kph on average. Therefore, the range of a fully tanked HFCV vehicle will be approximately 400 kilometres. With the assumption of a fast filling station, the refueling time per fuel unit is 37.4 seconds. Hence, filling up an empty tank will take roughly 20 minutes. Acquisition costs of a single truck (f ) and the costs of opening a HRS (U ) are rough estimates. The mentioned costs numbers are only for modelling purposes. A elaborated analysis on these parameters will follow in the results sec-tion.

Table 6: Parameters - Scenario 1

Parameters Value Y 32,09 g 0,62 r 0,08 Q 100% f €150.000 U €1.000.000

The demand at each customer node are illustrated in Table 5. These numbers can be interpreted as percentages relating to the load capacity of a single vehicle (100%). For the route with the most supermarket visits, one single truck needs to plenish all supermarkets in the network. In order to get a re-alistic weighted demand across all nodes, the quotient of su-permarkets in a node and the total susu-permarkets are used to

estimate customer demand. The load capacity of the truck

(23)

The time frame considered in the model is 24 hours. Due to regional policy on city logistics

truck delivery is only allowed between 5 am and 12 am for Groningen and Leeuwarden. The service time at each supermarket is estimated to be 45 minutes. In this time span the truck is parked, unloaded and readied to depart again.

As already discussed, four potential hydrogen refueling locations are considered: Pesse, Delfzijl, Groningen and Heerenveen. These locations are indicated with a triangle in Figure 2. The costs of opening a refueling station is assumed to be equal for each of the four potential refueling locations. Moreover, the distribution centre is located at Beilen. This implies that both the begin and end depot are located in Beilen.

Figure 2: Network with HRS Locations

By changing the parameter of vehicle range, it is investigated for which vehicle range (i.e. vehicle fuel capacity or fuel consumption) HRSs are required to open. Furthermore the effect of changing load capacity, demand parameters and the vehicle & HRS costs are evaluated.

(24)

4.2.2 Scenario 2 - Parcel Delivery

In the second scenario a parcel delivery company is considered. Similar to scenario 1, the network presented in Figure 1 and 2 is regarded. The company delivers packages to all customer nodes in the HEAVENN region. It is assumed that the company delivers packages to one central drop-off point where customers or any other downstream distribution vehicles pick-up the packages. The start and end depot is considered to be in Heerenveen. From there, packages are delivered to drop-off points across the region. One trip is considered for which the HFCV has to deliver to all demand nodes. Similar to scenario 1, the goal is to evaluate whether the transition from fossil fuelled trucks to HFCVs is feasible from the perspective of the business owner.

Table 7: Parcel Delivery Demand

Node City Population Demand ρ

Beilen 9656 1% Assen 67963 9% Groningen 231618 30% Delfzijl 46051 6% Drachten 45507 6% Heerenveen 50257 6% Leeuwarden 92695 12% Sneek 35534 5% Meppel 33564 4% Hoogeveen 55662 7% Emmen 107775 14% Total 776282 100%

Table 8: Parameters - Scenario 2

Parameters Value Y 32,09 g 0,62 r 0,08 Q 100% f €150.000 U €1.000.000

Input Data Scenario 2

Similar to scenario 1, the Hyundai XCIENT HFCV is

assumed to be deployed. The demand for parcel

de-livery is estimated using population size. The

popula-tion size for each city in the network has been expressed as a percentage of the sum of all population considered

in the network. The load capacity is considered to be

100% and equal to the sum of population percentages

(Ta-ble 7). With unlimited vehicle range this would imply

that one vehicle would be sufficient to deliver all pack-ages.

(25)

The main differences between scenario 1 & 2 are the location of the start/end depot and the demand at each node in the network. For both scenarios all nodes have to be visited. As a result of dissim-ilar demand the absolute amount of trucks required to replace the conventional vehicles is different in both scenarios. This leads to a difference in capital costs for each scenario and consequently impact the business case. As a result of the different depot locations in each scenario, the amount of kilometers travelled in the network differs, as vehicles have to return to a different node. This leads to differences in the total fuel consumption. How these aspects affect the business case is evaluated in the results.

4.2.3 Scenario Justification

The reason to consider these two specific businesses is fourfold. First of all, the logistics concerned with these businesses are of such size that genuine impact can be made in decarbonizing HGV transport in the northern Netherlands. Second of all, there are several similar businesses in the region for which the analysis of the feasibility of the transition to HFCVs can be replicated. Third of all, a typical feature for parcel delivery and supermarket resupply is that for both services trucks often have to deliver to many different locations. Intuitively, parcel delivery and supermarket resupply might thus be more complex to transition to HFCVs, as HRS infrastructure is limited.

Therefore these scenarios make a good case-study. Fourth and last of all, the business under

(26)

J.W.Hoekman 5 Results

5 Results

The following section outlines the result obtained from the model analysis and the insights from the business case. First, the model results of each individual scenario are discussed, providing the routing and locations decisions for a supermarket logistics and a parcel delivery company. Second, the business case-related aspects are discussed, concluding with the Total Cost of Ownership (TCO) for both scenarios.

The results were obtained with use of the Python programming language and Gurobi 9.0.1 op-timization software (Academic License) executed on a Acer Swift (CPU: 2.7 GHz, RAM: 8 GB).

5.1 Results per Scenario

This section discusses the results regarding locations and routing decisions obtained from the model analysis for the two scenarios (supermarket logistics and parcel delivery services).

5.1.1 Results Scenario 1 - Supermarket Logistics

The parameters of scenario 1 presented in the previous section are applied to the model. The re-sults found for scenario 1 are global optimums. In the first scenario a distribution center of a large supermarket chain in the northern Netherlands is considered. The start and end depot are located in Beilen (node 0 & 1).

In Table 9 the decision variables are presented. These variables are given for one day. The bi-nary decision variable xk

ij indicates whether the vehicle travels from node i to j. For example, xk

9,10 = 1 indicates that vehicle k travels from node 9 to node 10. The vehicle routes are presented in Figure 3. The model employs 2 HFCVs to make the delivery to the supermarkets on daily basis. Truck 1 serves all supermarkets except the ones in Emmen (node 11). The supermarkets in Em-men are served by truck 2. The total vehicle acquisition costs sum up to €300.000 (€150.000 per vehicle). The total distance travelled by both vehicles is 506 kilometres.

Since no HRSs are required to open, the total costs of opening hydrogen stations in this scenario are zero. The vehicles can reach and satisfy all demand on one full tank (remember that it is assumed that vehicles start with a full fuel tank). The vehicle range is 400 kilometres (Yk_/rk_).

The current load of the vehicle (qk

i) decreases over time as trucks deliver their cargo to the su-permarkets. The same holds for the current fuel level (yk

(27)

J.W.Hoekman 5.1 Results per Scenario Table 9: Decision Variables Scenario 1

Truck 1 xk ij τik qki yik xk 0,2 1 τ2k 60 qk0 70 y0k 32,09 xk 2,4 1 τ4k 165 qk2 70 y2k 30,65 xk 4,3 1 τ3k 300 qk4 62 y4k 25,03 xk 3,5 1 τ5k 628 qk3 58 y3k 21,82 xk 5,7 1 τ7k 720 qk5 44 y5k 18,13 xk_7,8 1 τ₈k 1163 qk₇ 40 y₇k 13,72 xk_8,6 1 τ₆k 1230 qk₈ 14 y₈k 9,15 xk_6,9 1 τ₉k 1310 qk₆ 10 y₆k 7,06 xk_9,10 1 τ₁₀k 1378 qk₉ 6 y₉k 3,77 xk_10,1 1 τ₁k 1440 q₁₀k 2 y₀k 1,60 Truck 2 xk ij τik qki yik xk 0,11 1 τ11k 93 q11k 10 yk11 27,60 xk 11,1 1 τ1k 1440 qk1 0

The effect of different vehicle ranges on the routing and station locations decisions are evaluated by running the model for ranges varying from 100 km to 800 km. In Figure 4 the number of HRSs, HFCVs and the respective vehicle ranges are depicted. These graphs are based on HRS opening costs of €1.000.000 and acquisition costs of €150.000 (fk_/Ui _{= 0.15). It can be observed that for} lower vehicle ranges either more vehicles are required or HRSs need to open up to make sure that all demand nodes can be satisfied. First, vehicles are added to make sure the round-trip can be com-pleted. However, if this is no longer possible due to the limited range, an additional HRS is required. Moreover, the total distance travelled is given in Figure 5. It can be seen that the travelled distance increases when more HFCVs are deployed. This is quite obvious as each vehicle needs to travel from the depot to a customer and return to the depot. The peak in the graphs relates to the case when vehicles have a range of 250 km and 5 HFCVs are required to satisfy demand. For this case the total distance travelled will be 930 km.

(28)

J.W.Hoekman 5.1 Results per Scenario

Figure 3: Vehicle Routing Scenario 1

At a vehicle range of 237,5 km, the first HRS required to open up is HRS Heerenveen. For a range of 175 km, HRS Groningen is added. For the lowest feasible vehicle range (100 km), HRS Pesse needs to open up as well and thus a total of three HRSs are required.

(29)

Figure 4: Number of HRSs & HFCVs - Scenario 1

Figure 5: Total Distance Travelled - Scenario 1

5.1.2 Results Scenario 2 - Parcel Delivery

The parameters of scenario 2 presented in the previous section are applied to the model. In the second scenario a parcel delivery company is considered. The start end depot are located in Heeren-veen (node 1, 0 & 7).

(30)

J.W.Hoekman 5.1 Results per Scenario Table 10: Decision Variables Scenario 2

Truck 1 xk ij τik qki yki xk 0,10 1 τ10k 329 qk0 100 yk0 32,09 xk 10,11 1 τ11k 382 q10k 77 y10k 24,95 xk 11,12 1 τ12k 443 q11k 73 y11k 22,78 xk 12,2 1 τ2k 521 q12k 66 y12k 19,90 xk 2,3 1 τ3k 566 qk2 52 yk2 15,40 xk_3,5 1 τ₅k 656 qk₃ 51 yk₃ 13,96 xk_5,4 1 τ₄k 720 qk₅ 42 yk₅ 8,34 xk_4,6 1 τ₆k 784 qk₄ 36 yk₅ 5,13 xk_6,1 1 τ₁k 1440 qk₅ 6 yk₆ 1,93 Truck 2 xk ij τik qki yki xk 0,9 1 τ9k 52 qk0 100 yk0 32,09 xk 9,7 1 τ7k 663 qk9 100 yk9 30,00 xk 7,8 1 τ8k 720 qk7 18 yk7 4,97 xk_8,1 1 τ₁k 1440 qk₈ 12 yk₈ 2,49

1) is adjacent to only two other cities.

Again the costs of opening hydrogen stations are zero, as no HRS stations are opened. The two vehicles can reach and satisfy all demand on one full tank with a range of 400 kilometres (Yk/rk). The sensitivity of the vehicle range (100 - 800 km) is inspected for scenario 2 as well. In Figure 7 the number of HRSs, HFCVs and the respective vehicle ranges are depicted. Still, these graphs are based on HRS opening costs of €1.000.000 and acquisition costs of €150.000 (fk_/Ui _{= 0.15).} Similar to scenario 1 it is observed that for lower vehicle ranges either more vehicles are required or HRSs need to open up to make sure that all demand nodes can be satisfied. However for lower vehicle ranges, scenario 1 requires more HRSs or HFCVs than the parcel delivery case. This is due to the location of the depot of the parcel distribution center which is connected to more nodes directly than the supermarket distribution center. This can also be drawn from Figure 8. The total distance travelled is lower for each range compared to scenario 1. Again, for the case Ui _{= 0 all} HRSs will be opened and the total distance travelled will equal 466 for all examined ranges (100 -800km).

(31)

Figure 6: Vehicle Routing Scenario 2

(32)

Figure 7: Number of HRSs & HFCVs - Scenario 2

(33)

J.W.Hoekman 5.2 Business Case

5.2 Business Case

From the results of scenario 1 and 2, it can be concluded that HFCVs are able to deliver to all customer nodes in the network for both scenarios. Based on a truck range of 400 kilometres no HRSs are required to refuel the trucks along their route. However, it should be noted that at least one HRS is required to make sure the trucks have full fuel tanks at start.

For scenario 1 the most suitable location would be HRS Pesse as both trucks travel along this node back to the depot in Beilen. For scenario 2 HRS Heerenveen would be best, as this is the location for the depot as well. Before leaving or returning to the depot trucks should be refuelled to full capacity at these HRSs to make sure they never run out of fuel. The example business cases for the two scenarios are explored using the previously obtained results and cost analyses.

5.2.1 Cost Analysis

As already discussed before, both operational and financial factors influence the feasibility of the transition to HFCVs. The following section discusses the financial aspects of this transition for scenario 1 (supermarket logistics) and scenario 2 (parcel delivery services), providing an estimate of the TCO for both business owners. First, the total costs are estimated, which consist of CAPEX and operational costs. Second, the potential financial benefits of HFCVs are discussed.

In the cost analysis of both scenarios, no subsidies, taxes and labour costs are taken into account. Vehicles are used for the entire life-span and salvage values are assumed to be zero. Moreover, the development of the hydrogen price is incorporated in the analysis. Roest et al. (2020) [49] approximated the hydrogen selling price in the Netherlands to be €10 per kg. Also the production price of hydrogen as stated in the Grant Agreement of the Hydrogen project of €3 per kg (Grant Agreement HEAVENN, 2019) [6] is considered. For both of these two hydrogen prices the analysis is repeated. The costs analysis is provided in Table 11 & 12.

Capital Expenditures

The cost analysis consists of the CAPEX, which includes the acquisition costs of a fleet of HFCVs and the operational costs of operating the fleet. In order to calculate the total CAPEX of a fleet of HFCVs it is necessary to know how many trucks are required on daily and yearly basis. The required number of trucks on trip basis is already obtained from the previous section and is used to get to daily and yearly numbers.

(34)

trip replenishing all supermarkets), a total of 70 trucks are needed on daily basis to replace all conventional trucks.

Table 11: Total Cost of Ownership - Scenario 1: Supermarket Logistics

Cost Analysis Scenario 1 - Supermarket Logistics

CAPEX Trip Daily Yearly

Acquisition Optimistic € 300.000 € 10.500.000 € 10.500.000

Current € 600.000 € 21.000.000 € 21.000.000

Costs Trip Daily Yearly

Operational Costs Fuel € 406 € 28.416 € 10.371.729

Margin HRS € 41 € 2.842 € 1.037.173 Maintenance € 29 € 2.054 € 749.841 Others expenses € 14 € 474 € 172.974 Total Costs € 489 € 33.786 € 12.331.716 Total Travelled Km 506 35.420 12.928.300 Operational costs/km € 0,95 TCO/km (Optimistic) € 1,77 TCO/km (Current) € 2,58

Table 12: Total Cost of Ownership - Scenario 2: Parcel Delivery

Cost Analysis Scenario 2 - Parcel Delivery

CAPEX Trip Daily Yearly

Acquisition Optimistic € 300.000 € 900.000 € 900.000

Current € 600.000 € 1.800.000 € 1.800.000

Costs Trip Daily Yearly

Operational Costs Fuel € 374 € 2.243 € 818.728

Margin HRS € 37 € 224 € 81.873 Maintenance € 27 € 162 € 59.191 Others expenses € 14 € 81 € 29.653 Total Costs € 452 € 2.711 € 989.445 Total Travelled Km 466 2.796 1.020.540 Operational costs/km € 0,97 TCO/km (Optimistic) € 1,85 TCO/km (Current) € 2,73

For scenario 2, it is assumed trucks operate 24 hours a day. On average, three trips a day are needed to deliver all packages (following the routing of Figure 4). For each of these trips, two vehicles are required to deliver parcels to each distribution centre. This implies that on a daily basis, 6 trucks are required to deliver all parcels in the northern Netherlands.

(35)

Operational Costs

The operational costs can be split into fuel costs, HRS margin, maintenance costs and other ex-penses.The fuel costs are calculated using the total fuel required (hydrogen/kg) to travel the total distance travelled multiplied with the hydrogen prices. furthermore, the gross margin for using the HRS station is considered. This is estimated to be 10%. The costs of opening a HRS station are not considered in the cost analysis because the business owners are consumers of the HRSs. Refueling station owners will have to consider these costs in their business case.

Moreover, it is assumed that maintenance costs are a fixed amount per kilometre and thus do not change over time. Even though hydrogen trucks have less maintenance intensive parts (such as rotating parts, gears) than diesel trucks, the price of replacements parts are higher. Maintenance costs are therefore assumed to be €0.058/km, similar to diesel trucks (Zhou et al., 2017) [50]. The other expenses include general costs such as insurances and vehicle registration costs. Based on the study of Konda et al. (2011) [51], these costs are estimated to be €6.77/day/HFCV (corrected for currency).

A similar cost of ownership calculation for a hydrogen price of €3 per kg can be found in

Ap-pendix A.3 & A.4. Conventional Trucks

In order to compare the costs of HFCVs to the current fleet, a similar cost analysis is made for conventional diesel trucks (Table 13). In this calculation a diesel price of €1,34 per liter is used. This price includes the refueling station margin. The acquisition costs of a new diesel truck are assumed to be €110.000. The vehicle range is assumed to be 900 km. Due to this range, only one vehicle is required per trip in both scenarios. This results in a fewer total travelled distance for conventional trucks. The total cost of ownership (TCO) comparison can be found in Table 14.

Table 13: Total Cost of Ownership - Conventional Diesel Truck

Cost Analysis Scenario 1 - Supermarket Logistics Scenario 2 - Parcel Delivery

CAPEX Trip Daily Yearly Trip Daily Yearly

(36)

J.W.Hoekman 5.2 Business Case Table 14: TCO - Comparison

Hydrogen selling price €10/kg €3/kg

Acquisition Costs Scenario 1 Scenario 2 Scenario 1 Scenario 2

Current: HFCV f = €300.000 €2,58 €2,73 €1,96 €2,12 Optimistic: HFCV f = €150.000 €1,77 €1,85 €1,15 €1,23 Conventional Truck: f = €110.000 €0,80 €1,12 €0,80 €1,12

From Table 14 it can be seen that for both scenarios and both hydrogen prices the conventional diesel truck has the lowest costs per kilometre. The drop of 7 euros in hydrogen selling price is not enough for the HFCVs to go up against the TCO of conventional trucks. Even with optimistic acquisition costs, a gap remains between the TCO of diesel & HFCVs.

5.2.2 Benefits

For both scenarios no financial benefits resulting from the transition are taken into account. How-ever, there may be benefits from operating a ”green fleet of HFCVs” that can move the TCO for HFCV’s closer to the TCO for conventional vehicles.

An important instrument in enhancing the business case for fuel cell vehicles are subsidies. In the Netherlands, for example, businesses are able to receive a subsidy for purchasing hydrogen ve-hicles. This obviously reduces the CAPEX costs for the business owner and may reduce the TCO for HFCVs. For the scenarios in this research, operating a fleet of conventional vehicles is more economical than a fleet of HFCVs. Hence, subsidies are required to stimulate HFCV adoption for these business. If hydrogen prices are €3/kg, a large subsidy on HFCV purchase is able to bring the TCO of HFCVs to the level of conventional vehicles. However, if the hydrogen price is€10/kg, a subsidy on HFCV purchase will not be sufficient, as the operational costs per km are already higher than the TCO of the conventional vehicles in both scenarios. In the Netherlands, the order of magnitude of this subsidy is generally thousands of euros and would be insufficient for the busi-nesses under consideration. Hence, higher levels of subsidizing are required if governments want to stimulate HFCV adoption.

Another instrument used by governments are taxes. For instance, there are concrete plans by the Dutch government to introduce a carbon tax in the Netherlands. This carbon tax will be aligned with the EU-ETS system, in which a market price for carbon emissions is created. The Dutch carbon tax will be the difference between a statutory carbon rate (set by the government) and the fluctuating market price. Hence, this tax will result in increased costs for businesses that

emit carbon. Consequently, the TCO of CO2 emitting conventional trucks may increase and make

(37)

J.W.Hoekman 5.3 General Results

Lastly, it could be that businesses are able to obtain value from the market if customers are willing to pay a fee for the transition to a zero-emission fleet. Most likely a combination of the three benefits will make HFCVs competitive with conventional vehicles.

5.3 General Results

From a network design perspective, the relation between vehicle acquisition costs and the costs of openings HRSs affect the optimal layout. If at a certain moment the costs of opening HRSs are equal or lower to the acquisition costs of a vehicle, the optimal (cost-minimal) infrastructure would consist of more HRSs. When more HRSs open, fewer vehicles are required to make the round-trip for a single business. Hence, there is a trade-off between the HRSs that are opened and the required fleet of HFCVs.

From operational perspective, it is possible for both businesses under consideration in this research to make the transition to a fleet of HFCVs. The results show that no HRSs are required if a fleet of HFCVs can be acquired with a vehicle range of at least 250 kilometres. With this range, both businesses are able to serve all demand nodes considered in the network without having to refuel. However, as vehicles require a full-tank at start, they should be able to refuel at least at one HRS somewhere. Currently, there are already some HRSs realized in the HEAVENN network and there are plans to realize many more. From an operational perspective, the tour planning as outlined in this research would enable businesses to replace conventional trucks with HFCVs.

The main differences between both businesses lie in the depot locations and demand at customer nodes. The results show that, if each customer node has to be visited, the demand at that specific node does not influence the routing of the vehicle. However, the location of the depot does influence the routing and the travelled distance: a depot location with multiple arcs that are adjacent to more customer nodes will result in fewer kilometres travelled. Businesses would do well to take this into account in the placement of their depots.

(38)

J.W.Hoekman 6 Discussion

6 Discussion

This section first describes the contributions of this study to the academic literature, as well as the practical implications for businesses. The second part of this section outlines the limitations of this research and provides recommendations for future research.

6.1 Implications

This research paper contributes to the academic literature in several ways. First of all, this paper is, to the knowledge of the author, the first to research the location routing problem in a real-life European case, focusing on heavy weight transport using hydrogen fuel cell vehicles. Most literature focuses either on model solution techniques, EVs or conventional vehicles or does not include the location routing problem. Therefore, one of the main contributions of this research is applying the hydrogen location routing problem model by Kamer (2017) [12] to an actual case. This specific model has never been tested before on real-life data. In applying the model to the case of heavy weight transport HFCVs, the aforementioned literature gap has been filled.

This paper provides several valuable insights for business. It is one of the first to consider the hydrogen refueling infrastructure of the HEAVENN region in a LRP model. The results show that, if transport-demand and -routes are consistent and plannable, this model can be used to assess the operational feasibility of transitioning from conventional vehicles to HFCVs heavy duty vehicles. When the HRS infrastructure becomes more developed, the model can be used as a decision support tool for businesses for strategic route planning and depot location placement.

Building on the model analysis in the previous section, it can be concluded that from an oper-ational perspective, both business under consideration in this research are able to make a transition to heavy duty HFCVs. However, from the business case evaluation it becomes clear that the fi-nancial feasibility remains equivocal. This ambiguity stems from the uncertain development of hydrogen prices, acquisition costs and governmental interference by means of taxes and subsidies.

6.2 Limitations & Directions for Future Research

(39)

J.W.Hoekman 6.2 Limitations & Directions for Future Research

Despite having constructed realistic data estimates (based on population size and facility den-sity) for the scenarios under consideration, it would further improve the accuracy of the output if regional data on the travel figures with a higher level of accuracy are employed in the analysis. This might require interviewing regional businesses or measuring HGV traffic flows. To be able to incorporate the relation between fuel consumption and vehicle weight, data is also required on freight weights and truck capacities. Furthermore, for the network under consideration only four potential refueling station locations were taken into account. In the future there might be increased interest in opening HRSs at different locations. It is relevant to explore which other locations these might be and how HRS utilization through potential usage by other business modes can be improved. Moreover, the capacity of HRSs has not been taken into consideration. Therefore, it is not possible to evaluate whether HRSs are actually able to meet the demand for hydrogen. This is especially important when more HFCVs are being adopted and the demand for hydrogen increases. In order to develop the HRS infrastructure, HRS owners need to know how HRS capacity is required to expand in order to meet the increasing demand of hydrogen. Extending the model with station capacity or even the local production of hydrogen would make a great improvement to the model. This could be done by adopting a similar approach as Bruglieri et al. (2019) [52] by modeling a capacity constraint on the HRSs. Furthermore, future research should focus on the business cases from the perspective of HRSs owners.

(40)

J.W.Hoekman 7 Conclusion

7 Conclusion

This study aimed to answer the following research question: What are optimal hydrogen refueling station locations and optimal routes for heavy goods vehicles in the northern Netherlands?

The answer to this question was obtained through application of a mixed integer linear programming model from Kamer (2017) [12] based on the location routing problem considering both locations and routing decisions on a real-life case in the northern Netherlands.

The majority of prior research either focuses on model solution techniques, EVs or conventional vehicles, or does not include the location routing problem. This paper is, to the knowledge of the author, the first to research the location routing problem in a real-life European case, focusing on heavy weight transport using hydrogen fuel cell vehicles. This paper thus fills a gap in the current literature by applying the MILP model to the HEAVENN case of transitioning to heavy weight transport HFCVs.

The MILP model was applied using realistic data for two business scenarios: supermarket logistics and parcel delivery services. The optimal routing and HRS locations for these businesses have been evaluated. It can be concluded that for the businesses under consideration, HFCV fleet adoption is feasible from an operational perspective: all considered trips could be made without being impeded by the current HRS infrastructure.

Moreover, the business case of transitioning to HFCVs was explored. From a financial perspective, the total costs of ownership for a fleet of HFCVs are currently higher than for a fleet of conven-tional vehicles. The future development of the TCO will primarily depend on the development of hydrogen prices, acquisition costs and governmental interference through taxes and subsidies. This development will influence the financial attractiveness of making the transition to heavy weight HFCVs.

(41)

J.W.Hoekman REFERENCES

References

[1] K. Adamson and P. Pearson. Hydrogen and methanol: a comparison of safety, economics, efficiencies and emissions. Journal of Power Sources, 86:548–555, 2004.

[2] U.S. Department of Energy. Fuel cell technologies program. page 1–2, 2011.

[3] M. Moultak, N. Lutsey, and D. Hall. Transitioning to zero-emission heavy-duty freight vehicles. The International Council on Clean Transportation, 53, 2017.

[4] Deloitte. Fueling the future of mobility- hydrogen and fuel cell solutions for transportation. Executive Summary, 1, 2020.

[5] C. Zhang, J. B. Greenblatt, M. Wei, J. Eichman, S. Saxena, M. Muratori, and O. J. Guerra. Flexible grid-based electrolysis hydrogen production for fuel cell vehicles reduces costs and greenhouse gas emissions. Applied Energy, 278:115651., 2020.

[6] New Energy Coalition. Grant agreement heavenn. 2019.

[7] M. W. Melaina. Initiating hydrogen infrastructures: Preliminary analysis of a sufficient number of initial hydrogen stations in the US. International Journal of Hydrogen Energy, 28(7), 2003. [8] M. Kuby and S. Lim. The flow-refueling location problem for alternative-fuel vehicles.

Socio-Economic Planning Sciences, 39(2):125–145, 2004.

[9] S. Erdogan and E. Miller-hooks. A green vehicle routing problem. Transportation Research Part E: Logistics and Transportation Review, 48:100–114, 2011.

[10] S. Salhi and G. K. Rand. The effect of ignoring routes when locating depots. European Journal of Operational Research, 39(2):150–156, 1989.

[11] M. Schwardt and K. Fischer. Combined location-routing problems-a neural network approach. Annals of Operations Research, 167(1):253–269, 2009.

[12] W. Kamer. Hydrogen fleet size and mix location routing problem with partial recharging. Working Paper, 2017.

[13] P. Kluschke, R. Nugroho, T. Gnann, P. Pl¨otz, M. Wietschel, and M. Reuter-Oppermann.

Optimal development of alternative fuel station networks considering node capacity restrictions. Transportation Research Part D: Transport and Environment, 78:102189, 2020.

[14] D. Baek, Y. Chen, N. Chang, E. Macii, and M. Poncino. Battery-aware electric truck delivery route exploration. Energies, 13(8):1–18, 2020.

(42)

J.W.Hoekman REFERENCES

[16] G. B. Dantzig and J. H. Ramser. The truck dispatching problem. Management Science, pages 80–92, 1958.

[17] R. G. Conrad. The recharging vehicle routing problem. IIE Annual Conference, Proceedings, pages 1–8, 2011.

[18] W. Zhang, Y. Gajpal, S. S. Appadoo, and Q. Wei. Multi-depot green vehicle routing problem to minimize carbon emissions. Sustainability, 12(8), 2020.

[19] M. Schneider, A. Stenger, D. Goeke, M. Schneider, A. Stenger, and D. Goeke. The electric vehicle-routing problem with time windows and recharging stations. Transportation Science, 48(4):500–520, 2014.

[20] G. Hiermann, J. Puchinger, S. Ropke, and R. F. Hartl. The electric fleet size and mix vehicle routing problem with time windows and recharging stations. European Journal of Operational Research, 252(3):995–1018, 2016.

[21] H. Kim, M. Eom, and B. I. Kim. Development of strategic hydrogen refueling station deploy-ment plan for Korea. International Journal of Hydrogen Energy, 45(38):19900–19911, 2020. [22] O. Berman, R. C. Larson, and N. Fouska. Optimal location of discretionary service facilities.

Transportation Science, 26(3):201–211, 1992.

[23] M. J. Hodgson. A flow-capturing location-allocation model. Geographical Analysis, 22(3):270– 279, 1990.

[24] P. Jochem, C. Brendel, M. Reuter-Oppermann, W. Fichtner, and S. Nickel. Optimizing the allocation of fast charging infrastructure along the German autobahn. Journal of Business Economics, 86(5):513–535, 2016.

[25] S. Lim and M. Kuby. Heuristic algorithms for siting alternative-fuel stations using the Flow-Refueling Location Model. European Journal of Operational Research, 204(1):51–61, 2010.

[26] I. Capar and M. Kuby. An efficient formulation of the flow refueling location model for

alternative-fuel stations. IIE Transactions, 44(8):622–636, 2011.

[27] S. A. MirHassani and R. Ebrazi. A flexible reformulation of the refueling station location problem. Transportation Science, 47(4):617–628, 2013.

[28] I. Capar, M. Kuby, V. J. Leon, and Y. J. Tsai. An arc cover-path-cover formulation and strate-gic analysis of alternative-fuel station locations. European Journal of Operational Research, 227(1):142–151, 2013.

Optimizing Hydrogen Refueling Station Locations and Tour Planning: Heavy Weight Road Transport in the Northern Netherlands

Optimizing Hydrogen Refueling Station Locations

and Tour Planning:

Heavy Weight Road Transport in the Northern

Netherlands

Contents

Optimizing Hydrogen Refueling Station Locations and

Tour Planning:

Heavy Weight Road Transport in the Northern

Netherlands

Joris Hoekman

1

Introduction

2

Theoretical Background

2.1

Vehicle Routing Problem

2.2

Facility Location Problem

2.3

Location-Routing Problem

2.4

Contributions

3

Methodology

3.1

Model Formulation

3.2

Data Collection

4

Scenario Analysis

4.1

HEAVENN Project

4.2

Scenario Formulation

5

Results

5.1

Results per Scenario

5.2

Business Case

5.3

General Results

6

Discussion

6.1

Implications

6.2

Limitations & Directions for Future Research

7

Conclusion

References