下载掌阅APP，畅读海量书库
立即打开
Supply chain management is concerned with the management and integration of the various organizations involved in the upstream and downstream of the supply chain for enabling the better flow of products, services, finances and information from the primary sources to the final customers (Christopher, 1992, Chopra and Meindl, 2007). Because the supply chain system often involves a large network of organizations and a broad scope of activities to manage, making decisions in this complex environment can be very challenging, and there is therefore a need for appropriate methodologies to support this decision making.
Enabled by the recent advances in computing power, simulation modelling is increasingly becoming a viable approach to assist decision-makers to analyse the complex issues in supply chains. Simulation is an experimental approach used to mimic real-world systems that, unlike real-world systems, can then be experimented upon without being confronted with real-world consequences (Pidd, 2006, Peck,2004, Winsberg, 2003). The use of simulation modelling as a tool for researching supply chain theories and evaluating supply chain strategies has attracted growing attention. Various simulation techniques have been used to model a wide spectrum of supply chain issues at strategic, operational and tactical levels. In this context, one recently popular simulation modelling technique is Agent-Based Simulation (ABS).ABS is particularly well suited for modelling the behaviour of complex systems over time, and is thus potentially applicable to complex and dynamically developing supply chain systems.
Since a key focus of this research is to study the dynamic behaviours and effects of LHC, a simulation modelling approach needs to be employed for the effective analysis. This study considers Agent-Based Simulation (ABS) as a suitable modelling approach to explore the process of collaboration and how it links to performance behaviours. From the methodological point of departure, a review of the key characteristics and use of ABS modelling would be useful to inform the model design for studying LHC.
Given the fact that the prior research in the field of LHC has mainly been empirical and qualitative methods, there wasn’t in the past a real case of ABS research identified in this area. This part of the literature review therefore, extends to a broader scope to explore the ABS application in the wider supply chain management (SCM) context.
In the supply chain modelling literature, the ABS approach has been highlighted as an increasingly suitable method to model supply chain problems and its application has grown rapidly in recent years. The section hence explores the features and advantages of ABS, and how ABS has been applied to study issues in the SCM context. The rest of this section is structured as follows. Section 2.4.1 provides a brief review of the main features of the present supply chain modelling and simulation methods. In Section 2.4.2, the paradigm of ABS for modelling the supply chain is analysed and compared to other key supply chain simulation techniques. Section 2.4.3 explores how ABS has been applied to model some of the key issues in supply chain management.
The literature on supply chain modelling is vast and covers many approaches.Broadly speaking, current modelling methods can be grouped into two mainstream families: analytical modelling and simulation modelling (Gokhale and Trivedi, 1998). Analytical (or mathematical programming) modelling methods call for the solution of a mathematical problem using various algorithms and equations. Many Operational Research-based (OR) models belong to this family, including methods such as linear/mixed integer programming, queueing theory, game theory and network optimization models (Poler et al. 2013, Ravindran, 2008). On the other hand, simulation modelling methods are developed to mimic the behaviours of the various elements in a supply chain system over time in order to study and predict the outcomes from sample histories by running a simulation program (Altiok and Melamed, 2010).
It is argued by some authors that using analytical (or mathematical) modelling methods to model the supply chain could be less efficient because mathematical models are usually difficult to construct for realistic cases and can only tackle small scale problems (Thierry et al. 2010, Ahn and Lee, 2004). In contrast, simulation modelling methods tend to be more flexible for exploring the behaviours and performance of the large-scale situations that can exist in a typical supply chain system. A number of authors argue that there are several advantages of applying simulation over analytical modelling approaches when seeking to model the supply chain (Lee and Kim, 2008, Gokhale and Trivedi, 1998, Thierry et al. 2010, Chan and Chan, 2010, Nikolopoulou and Ierapetritou, 2012b, Altiok and Melamed, 2010). The advantages are:
● Analytical models rely strongly on assumptions that tend to be over-simplified. It is challenging to construct a model using mathematical programming when dealing with many variables (the so-called “curse of dimensionality”). Mathematical models are therefore usually limited to solving small-scale problems in the supply chain. Simulation models on the other hand, can capture and model many elements and variables. The assumptions for the modelling can be relaxed or made more realistic. The advantage of simulation over analytical modelling lies in the fact that very large numbers of detailed behaviours in the supply chain can be captured allowing the supply chain to be modelled from a broader perspective.
● Analytical methods seek optimized and exact solutions from the modelling, which requires the problems to be formulated in a static and deterministic way, which is difficult to apply when facing a complex and stochastic environment. In contrast, simulation methods seek good but approximate solutions. The ability to carry out “what-if ” configurations provides additional flexibility to identify a “best” configuration, which further strengthens the adoption of this approach.
● As opposed to the analytical models, which are built on abstract mathematical expressions, simulation models often use representations that are closer to reality. Many simulation tools support the graphical visualization of the system behaviours through animation, which helps the modeller and model users to observe the temporal evolution of the model’s state and statistics in the course of a run. Such features facilitate a better understanding of the modelled system.
Overall, when the finer level of detail and broader model boundary are important,the advantages of simulation models outstrip those of analytical models. Simulation modelling is therefore argued to be a more effective approach for modelling supply chain systems that are usually large-scale and complex in nature.
Supply chain system modelling is challenging due to the broad scope of issues and complex interactions between supply chain organizations. In recent years ABS has been increasingly recognized to be suitable for modelling supply chain problems due to its ability to model the complexity of the supply chain system. Before ABS, two classic simulation approaches were widely used for supply chain modelling, namely Discrete-Event Simulation (DES) and System Dynamics (SD).
This section compares the key characteristics of these three approaches for modelling the supply chain system in order to distinguish ABS from the classical supply chain simulation approaches, highlighting the advantages of the former.
DES models suggest that real-world systems and processes are represented by a set of distinct events (Altiok and Melamed, 2010). From a technical standpoint, DES views a system as a network of activities and queues where state changes occur at discrete points of time (Brailsford and Hilton, 2001). In DES, objects and people are modelled individually and can be referred to under the generic term of “entities”.Figure 2.1 demonstrates a classic DES approach to modelling the design of a supply chain. In this Figure, raw material entities enter the circuit board manufacturing process and are transformed into circuit board entities. Next, the circuit board entities enter the next supply chain tier, i.e. the Surface-Mount Technology (SMT) manufacturing process. Finally, the entities leave SMT as the final product entities.
Figure 2.1 Example of a discrete event simulation model of a supply chain design
In the literature, it is generally argued that the DES approach is more suitable for modelling Supply Chain Management (SCM) issues at the operational level(Lane, 2000, Sweetser, 1999, Taylor and Lane, 1998), since it is more focused on the process of the supply chain activities and an individual entity’s journey through the modelled supply chain. Hence, DES models are most popular for modelling supply chain problems such as network configurations, inventory control policies,manufacturing planning and scheduling related to the queuing problems. From DES perspective, the elements that describe the structure of the supply chain (events,activities and processes) are considered as passive “objects” that are pre-defined by the modeller. Chatfield et al. (2007) and Siebers et al. (2010) argue that using DES to model a supply chain usually implies a network perspective and focuses on representing the supply chain’s topology and infrastructure, while generally discounting the control and decision processes that occur within each supply chain player.
SD has its roots in Jay W. Forrester’s Industrial Dynamics (Forrester, 1961),which is a simulation approach that investigates the effect of information feedback and delays on the dynamic behaviours of a system. In the context of supply chain systems, the dynamics existing between firms in supply chains can cause errors,inaccuracies and volatility, hence creating huge uncertainties, which increase for operations further upstream in the supply chain (Slack et al. 2006). SD simulation is therefore a means of inferring the time evolutionary dynamics endogenously created by such system structures (Lane, 1997).
SD models represent a system as a set of stocks and flows (Brailsford and Hilton,2001, Pidd, 2006). Figure 2.2 shows an example of an SD model that captures the relationship between two stocks: retailer inventory and supplier orders. The retailer inventory increases with incoming shipments from its supplier and decreases with sales. Likewise, the supplier orders increase with incoming orders from the retailer and decrease with production which will then be shipped to the retailer.
Figure 2.2 Example of a system dynamics model of a supply chain design
In contrast to DES models, in SD models, the individual entities are not specifically modelled, but are instead collectively represented as a continuous quantity in a stock. Movement to or from a stock is represented by a flow, which is defined to be the rate of change of a stock. Another difference from DES is the management of time, in that the state changes are continuously monitored over time in SD models(the time is usually advanced by small discrete steps of equal length) (Angerhofer and Angelides, 2000, Akkermans and Dellaert, 2005). In addition, SD models are generally deterministic and variables usually represent average values (Pidd, 2006).
Due to the fact that SD models are usually built with a “distant” perspective, it has been claimed that SD is more suited to modelling the supply chain problems at a strategic level (Lane, 2000, Sweetser, 1999, Taylor and Lane, 1998). For instance, the typical supply chain problems that are frequently modelled using SD are supply chain integration and the bullwhip effect. The SD model approach has several advantages in respect to modelling the supply chain (Angerhofer and Angelides, 2000, Tako and Robinson, 2012). First, it cares about the structure of the system, which enables the modeller to take a holistic view of the supply chain system, integrating many sub systems. Second, the use of causal loop diagrams helps to capture a dynamic view of the cause and effect relationships between policies and decisions among the supply chain organizations. Third, building SD models requires less detailed data than DES models. On the other hand, the SD approach requires the modeller to represent the supply chain as a set of closed-form equations, which are inflexible when it comes to constructing the desired form of model. Also, the models do not represent supply chain processes that contain multiple stages, since the behaviours of individual supply chain participants are not explicitly modelled (Chatfield et al. 2007). SD, therefore,is not as effective as DES in modelling the operational issues in the supply chain,although it can be used for the early/intermediate stages of analysis and decision making when less detailed models or results are required.
ABS has been increasingly used for studying supply chain issues since it caters for the disadvantages inherent in the DES (discounting the control and decision process of individuals) and SD approaches (systems only profiled at the aggregated level). The ABS modelling paradigm focuses on modelling the individuals in the(supply chain) system, known as “Agents” who can represent people, machines or companies. Agents have autonomous behaviours, often described by simple rules, and interact with other agents, who in turn influence their own decisions and behaviours. The global (system-level) behaviours then emerge as a result of these myriad interactions of agents and their individual behaviours (Eppstein et al.2011, Niazi and Hussain, 2011, Macal and North, 2010). Since ABS centres on the individual characteristics and behaviours, it provides an effective way to study a system by explicitly modelling every unique individual of that system. A typical agent structure is illustrated in Figure 3. Agents interact with other agents in the supply chain network (e.g. place an order or fulfil an order). Based on the knowledge gained from the interaction, agents may change their attributes (e.g. inventory levels) and update their beliefs or behaviours (e.g. supplier selection rules).
Figure 2.3 Example of an agent-based model of a supply chain design
ABS has some specific features that have allowed it to be increasingly recognized as an alternative approach to modelling supply chain issues. First, the supply chain system is a complex system consisting of many individual organizations with different objectives and action strategies. It is easier to model these heterogeneous agents in ABS than in DES (due to the passive/few interactions between objects) and impossible in SD (since this assumes a homogeneous collection of individuals).In ABS, heterogeneity can be defined both for the agent characteristics and his decision/action rules. Second, learning and adaptive capability can be modelled for the individual agents. Such proactivity in responding to the changing environment is difficult to represent in DES models because of their fixed structure and process,but this is an important aspect in modelling the supply chain, where an explicit representation of human/organizational decision making is required in order to examine the system behaviours effectively. SD models can be used to model learning and adaptive capability at the population level but not at the individual level.
Consequently, ABS is an ideal approach for developing a model that requires the explicit analysis and representation of individual behaviours (such as inventory decisions on the part of every supply chain player), and it can also be used to examine the linkages between those micro agent details and the macro system behaviours (such as system costs and the bullwhip effect). From this viewpoint,employing ABS would help the modeller to obtain greater flexibility in terms of addressing both the macro and micro supply chain problems at the same time in the same model. Supply chain issues that can be effectively modelled by SD but not by DES (e.g. coordinated decision making) as well as those that are more suitable for DES than SD (e.g. manufacturing scheduling/the queuing problem) can both be integrated within an ABS model to form a hybrid model (Eldabi et al. 2016). In summary, ABS complements DES and SD as a simulation approach in supply chain modelling.
This paper seeks to provide a review of the ABS methods used in supply chain applications. The study is based on a review of the published papers that describe the modelling of supply chain issues. The research process for this review followed the procedures explained further below.
In order to identify the scientific literature that describes the use of ABS to study supply chain problems, a search was performed in the well-known journal databases(Business Source Premier (EBSCO), ProQuest , ABI, Emerald, Web of Science(WOS), Science Direct, IEEE Xplore, and ACM digital), looking for a combination of terms “agent-based simulation/modelling/model” and “supply chain” within paper titles, keywords and abstracts. The search considered only digital works written in English up to March 2015. Both academic journal papers and conference papers were included. Other publications such as books, trade papers, technical reports or newsletters were excluded. The search presented over 300 papers.
To ensure the identification of papers addressing the defined topic, a screening process was carried out. First, duplicate papers found from the different databases were removed, reducing the number of papers down to 116. Second, a quick scan of the abstracts of these 116 papers was conducted to filter out irrelevant papers, as well as papers with only the abstract available in each search engine. Following this scanning check a total of 91 papers were retained.
After the paper list had been retrieved, the full papers were downloaded. Since there could still be some irrelevant papers, an additional screening process was conducted. First, those papers that only described the conceptual model of ABS without any actual model implementation and experiments were excluded from the review list. The study aims to investigate how ABS is used to model supply chain issues hence the included articles must feature a complete ABS model. Second,those articles adopting the terminology “multi-agent system (MAS)” were examined carefully to distinguish the multi-agent system engineering problems from simulation studies using multiple agents. The latter is included because it is essentially the same as ABS. Finally, papers in which the keyword “supply chain” appeared in their titles or abstracts but which were not in fact focusing on supply chain problems were not considered. By applying these screening criteria, a final total of 73 relevant ABS papers was confirmed and used for the review analysis.
The literature shows that ABS has been used to study many types of supply chain problems. Table 2.1 shows the fields that ABS modelling has been applied to (the detailed breakdown can be found in the appendix). It is noteworthy to see among the list of supply chain issues that the most frequently modelled are related to the fields of supply chain collaboration, inventory management, risk/uncertainty management and supplier selection. This “big four” accounts for 77% of the total number of ABS applications and are therefore the main supply chain application fields using ABS.
Table 2.1 Agent-based modelling application fields in relation to supply chains
The following sections aim to provide a more detailed review of ABS applications within the four fields in order to understand the different ways in which the ABS approach has been applied and what results have been achieved.
The fundamentals of SCM are about creating a partnership to facilitate communication and collaboration between individual firms in the supply chain network.ABS seems to be particularly useful to model supply chain collaboration given that the ABS paradigm centres on individual behaviours and interactions, which enables the collaboration model to be profiled at a greater level of detail compared to other simulation methods. It also offers greater flexibility to model a wider range of different types of collaboration strategies. The literature shows that the collaboration within the supply chain in general can be modelled in two different forms: (1) collaboration based on sharing information; (2) collaboration based on sharing physical capacity and profits. The following subsections discuss the relevant applications in more detail.
Information sharing is regarded as a major form of collaboration in the literature. A supply chain is a complex network that involves a lot of local decisions and activities.As a result, none of the members in a supply chain can have a full picture of the networked operations, and consequently, face uncertainty when trading with each other in the network. This issue creates incentives for member parties (or agents) to pursue information sharing through collaboration in order for them to gain greater visibility of how others perform and thus better align their operations when they trade and collaborate. This shared information can be used by different supply chain agents to make wiser or more appropriate decisions when they operate as a part of a bigger supply chain network consisting of complex inter-organizational connections.
The literature revealed that information sharing in a supply chain context can be modelled in two main ways: sharing demand information and sharing supply information.
Sharing demand information between supply chain partners was found to be the most common. This includes specific demand information such as sales/order forecasts, point-of-sales (POS) data, and customer inventory depletion information.The following part presents some typical ABS works in order to illustrate how each type of demand information is shared and modelled, and what the effects are.
Caridi et al. (2005) studied the Collaborative Planning Forecasting and Replenishment (CPFR) process between a manufacturer and a retailer who are willing to collaborate in the exchange of sales and order forecasts. Three distinctive CPFR models were implemented, representing different levels of collaboration. The first model represents the conventional approach to CPFR, in which the retailer shares their order forecasts with manufacturers, whereas the manufacturers share their sales forecasts with retailers. Both then work together to try to narrow down the gap in their demand forecasts. The second (advanced) model enables trading agents a further collaboration ability to relax operational constraints for forecasting according to a priority list where the ranking for the relaxation of constraints is recorded. Within the third (learning) model, agents are more intelligent than in the previous ones, due to their ability to learn from the past, which allows them to collaborate to reset criteria threshold values (KPIs) for forecasting based on historical data that would indicate the product life cycle and market trend. Through modelling and comparison, it was concluded that CPFR strategies, coupled with dynamic constraints relaxation (i.e. the advanced model) and with criteria/rules updated through historical data analysis (i.e.the learning model), achieved greater benefits than the conventional CPFR in terms of total costs, inventory level, stock-out level and sales.
Xu and Zhu (2013) analysed the influence of demand information sharing in a retailer-dominant supply chain. Their model investigated four settings: (1) no information sharing; (2) information sharing between retailer and manufacturer;(3) information sharing between retailer and supplier; and (4) all the members of supply chain being involved in the information sharing. The simulation results showed that the total cost of the supply chain was the highest in the case of no information sharing. When demand information is accessed by all supply chain members, the total cost of the supply chain can be kept to a minimum. The result also suggests that sharing demand information has a much stronger value for the upstream players.
Bhattacharyya and Zhang (2010) examined the effect of demand information sharing between sellers and buyers in an E-commerce supply chain (B2B E-hub).Five different demand information sharing strategies were compared: (1) no information sharing; (2) sharing aggregated hub demand (AHD) information;(3) sharing aggregated end demand (AED) information; (4) sharing aggregated buyer inventory position (ABI) information; (5) a hybrid approach that combined AED and ABI. The simulation results showed that sharing both AHD and AED is beneficial to the sellers in terms of cost. This is mainly due to the lower inventory level and lower stock out penalty costs. While a lower inventory level results from a higher frequency of ordering, the same benefit is achieved with a lower order frequency(hence lower ordering cost) in ABI. This suggests that sharing the buyers’ inventory consumption information as the demand signal is more valuable than directly sharing the aggregated demand information. In addition, more does not necessarily mean better. Sharing more than one type of demand information together might not always be more beneficial as it might complicate the agent decision making in response to the demand changes.
Lin et al. (2002) examined the effects of demand information sharing on supply chain performance in electronic commerce. Three levels of information sharing were implemented in the model: order information, inventory information, and demand information. The findings indicated that sharing demand information achieved the lowest total cost, the highest order fulfilment rate, and the shortest order cycle time,whereas sharing order information leads to the lowest order fulfilment rate and the longest cycle time. When sharing buyers’ inventory depletion information, buyers tended to trade with a specific supplier for a longer period of time. When sharing order or demand information, however, buyers tended to switch suppliers more frequently.This demonstrated that sharing inventory information is a workable alternative when the cost of switching is relatively high.
Lau et al. (2004) investigated the impact of diferent levels of demand information sharing on supply chain performance under the various supply chain structures. Four levels of information sharing strategy were implemented, characterized by which supply chain echelons are engaged in the information sharing and the information type, such as sharing only the order information or sharing the mean and variance of demand with a purpose to hide the actual demand and cost structure. The results showed that no single level of information sharing dominates the others from the perspective of individual companies. However, the value of sharing demand information between downstream echelons is more significant in terms of supply chain operating costs than that of sharing information between upstream echelons,regardless of supply chain structures.
Chatfield et al. (2004) examined the effect of sharing demand forecast information. This information is used to predict the demand in lead-time and inventory parameters updating, which can affect the order streams, inventory levels and stock outs. With no information sharing, each supply chain node generates its own forecast based on their local information. The forecast is then used to generate the parameters for purchasing and supply. When information is shared, the nodes in the supply chain are aware of the current customer demand. This awareness informs their forecasts and allows them to fine-tune their planning parameters. The results showed that the sharing of downstream information decreased the demand variance amplification significantly for upstream players. This is because the end-customer’s demand order stream has a variance less than or equal to the variance of the orders coming from the downstream partner. This customer information smoothens the fluctuations in the planned inventory level so that the resulting order stream has a lower variance.Information sharing also protects a supply chain against “cascading failures” (stock outs), especially for a supply chain system with more echelons.
In contrast to sharing demand information, sharing the supply related information downstream to customers was considered only infrequently. The types of supply information that can be shared include capacity, inventory, backlog and lead-time.The following part introduces three available ABS works which modelled these kinds of information sharing.
Chan and Chan (2004) studied the retailer-supplier collaboration problem in terms of the quantity and delivery date flexibility needed to cope with the demand dynamics.Instead of setting a fixed delivery date, a bigger delivery window was allowed to ensure a more proactive collaboration between retailers and suppliers. Quantity flexibility, on the other hand, ensures that retailers can choose to receive a lower order quantity if their cycle demand was not as strong as predicted. Within the range of delivery dates,retailers and suppliers would enter into a collaboration process in which suppliers will repetitively check/share their latest inventory production status with retailers, who will then take this information from the supply side and decide if they want to ask their suppliers to arrange the order delivery earlier or later according to their own inventory depletion progress. Through the modelling, and comparison with the conventional order fulfilment approaches, where no inventory information sharing and flexible coordination are allowed, the collaboration model significantly reduced the total system cost and increased the order fill-rate (service level). Moreover, the proposed collaboration was not only able to reduce the total system cost/increase fill-rate, but the impact of increasing demand uncertainty was also suppressed. That means the marginal cost against uncertainty was reduced significantly.
Sawaya (2006) analysed the effects on supply chain performance of sharing suppliers’ lead-time forecasts with customers. Specifically, the supplier prepares an estimate of the internal queue time based on the current finished goods inventory position, current orders and backorders, the expected capacity in the future and the mean demand it expects in the future. Then the supplier gives customers their best guess of when orders it receives the next day will be shipped from the factory. This estimate is used by the customer in their calculations to determine their demand orders during lead-time in place of an estimate of the lead-time from observed orders and received shipments. The results suggested that, for the most part, the value of suppliers sharing shipment lead-time forecasts for future orders with their customers is not strong. Sharing lead-time information can have a net negative effect on the system and on the manufacturing organization since it appeared that this might create a mechanism that forces the supplier to keep greater inventory. When the variability of the lead-times is high and the demand volume is high, the sharing of lead-time information has greater benefits, however. This is even more pronounced when the capacity is variable and known somewhat in advance of the day on which the actual capacity is realized, which indicated that if the precision of the lead-time forecast is high enough, and then the benefits could be well worthwhile.
Ibrahim and Deghedi (2012) studied how the sharing of factory disruption information can help minimize the evolution of risk downstream in the supply chain and improve both the whole chain’s and each member’s performance. The model introduced factory breakdown as a source of supply disruption that can severely delay order delivery. When the factory does not notify the downstream elements of the supply chain of the breakdown problem, the downstream players only realize this problem when no shipment is received, and thus switch to another factory. In this scenario, all the entire downstream players will suffer from delayed shipments and the accumulation of backorders. When the factory immediately alerts them of the breakdown, however, the players throughout the whole chain can act swiftly to adjust their orders. The results showed a significant reduction in the cost of the supply chain and each of its agents due to the sharing of breakdown information. In addition, the analysis found that the significance of sharing breakdown information became greater with when the disruption frequency increased. Despite the possible confidentiality of the disruption problem to the factory agent, the results showed that the factory would benefit most in terms of cost reduction, if the breakdown information is shared.
Overall, the current ABS literature has shown that modelling the sharing of demand-related information upstream in the supply chain is more common than sharing supply related information to downstream supply chain organizations. This is perhaps driven by the conventional approach for managing the supply chain which is largely based upon demand forecasting and inventory planning. This indicates that the primary focus of modelling of information sharing is to make the demand associated data much more transparent and thus to give it more of a role in controlling the supply to match closely with demand (i.e. increasing forecast accuracy, reducing the bullwhip effect/excess inventory). Using ABS to model the sharing of supply information and using this to control and match the demand closer to the supply is only infrequently considered. In addition, comparing and combining the two types of information sharing represents an interesting implementation for ABS. There is certainly plenty of room for further studies in this regard.
Collaboration based on sharing physical capacity and profits are of high practical value and are often seen in many real collaborations in the supply chain. The literature review identified several ABS works that model these kinds of collaborations.
Albino et al. (2007) analysed the benefits of supply chain collaboration in industrial districts (ID). Collaboration was modelled in the form of sharing production capacity between similar firms (horizontal collaboration) at two supply chain stages (suppliers and buyers). The collaboration between suppliers seeks to balance the utilization of production capacity between suppliers, meaning that suppliers proportionally share the excess demand according to their available capacity thus ensuring their co existence in the long term. The collaboration between buyers, meanwhile, emphasizes the minimizing of the unsatisfied customer demand, meaning that unsatisfied orders tend to be allocated to the buyer agent with a higher level of available production capacity. Several demand scenarios and organizational structures were configured to represent the different ID supply chain environments for collaboration. The results showed that collaboration in production has a substantial positive effect on the ID performances in terms of efficiency and flexibility. Further, when the collaboration in IDs is characterized by the presence of leader firms, there is a greater improvement in efficiency at the expense of losing flexibility as the demand variability increases.
Xie and Chen (2005) studied the horizontal collaboration among retailers in the supply chain. This collaboration was among retailers who pursue the partnership for higher profits. The collaboration was modelled as full collaboration, i.e. when several retailers collaborate, they form a coalition and act as a single large retailer and they take a uniform price, and share their inventories, costs and profits. Under this configuration, all the retailers inside a coalition are highly coordinated. They all try to maximize the coalition’s total profit; yet different coalitions and outside retailers are purely competitive. The results revealed that the size of the collaboration network,and several other factors, can affect the stability of the coalition structure. When more participants are involved in the network, there are more incentives for partners to break from the existing partnerships. The collaboration also has prominent external effects, which makes it more beneficial for outside retailers.
Giannoccaro and Pontrandolfo (2009) studied a collaboration mechanism based on revenue sharing negotiation in the supply chain. The model assumed that the negotiation process between retailers and distributors is affected by three main variables, namely: the agent’s propensity to negotiate, the propensity to threaten to abandon the negotiation, and the propensity to collaborate. The simulation results showed that the best supply chain profits were obtained when the contractual power for both supply chain agents is low, regardless of the degree of collaboration. In such a case, both actors have a high propensity to negotiate and a low propensity to threaten to abandon the negotiation, and the negotiation therefore tends to end more frequently with an agreement. A high degree of collaboration for both supply chain agents also assures high supply chain profits, regardless of the contractual power. Thus, the best scenario is characterized by low contractual power and high collaboration for both agents. Further, it was found that the asymmetric distribution of contractual power between the actors reduces the chance of the adoption of revenue sharing.
From the above, ABS has shown a good ability to model various forms of supply chain collaboration. This is because ABS focuses on modelling the interactions and exchanges between agents which can incorporate the different sorts of shared materials.
The issues related to inventory management are found to be modelled intensively using simulation including ABS (Tako and Robinson, 2012). This literature review confirms this. For example, Zhang and Bhattacharyya (2010) examined and compared the inventory management and performance in the traditional and E-commerce supply chains. A typical order-up-to (OUT) inventory replenishment policy was implemented in the model for all supply chain agents to follow. The results identified that all agents tend to keep more inventories and backlog lose fewer orders in the e-marketplace than in traditional supply chains. The effects on upstream distributors and manufacturers are more profound than those on downstream retailers. Similarly, Dong et al. (2012) developed a model based on two continuous inventory replenishment strategies, the (R, S) and (Q, R) policies, to analyse the inventory replenishment performance in a three-stage supply chain. They found that under the (R, S) policy the service level (order fill-rate) is better, and the shortage costs are lower, but the inventory holding costs are slightly higher than those under the (Q, R) policy. Overall,the (R, S) policy is better than the (Q, R) policy as an inventory planning and control method. In another example, Moyaux et al. (2004) compared three ordering policies and their effectiveness in terms of reducing inventory variations and back order costs.
In addition to more traditional supply chain inventory models that focus on stable operational processes, the new features given by ABS simulation enable the modeller to configure the agent with a learning capability, allowing them to learn from past experiences so as to make better inventory decisions. For instance, Jiang and Sheng(2009) constructed an inventory model to investigate the dynamic inventory control issues under the non-stationary customer demand where the use of traditional time or event-trigger inventory policies were not accurate. Case-based reinforcement learning was applied and was proved experimentally to be effective in this situation.Similar benefits can also be found in a study by Kim et al. (2008). They developed an action-reward learning-based inventory control model for a two-stage serial supply chain with the non-stationary customer demand. Two learning strategies(centralized and decentralized learning) were implemented for comparison with the inventory control method with no learning. The results showed that the two learning models outperformed the inventory model without the learning control in terms of average inventory cost. In another application, Kim (2009) studied the effects of trust accumulation between supply chain trading agents. The learning capability was allocated to the agents, who could then use this to analyse the historical performance of their counterparts and hence increase or decrease their trust level towards each of them, thereby affecting the order and supply decisions. The simulation results revealed that agents’ decisions on forecasting, ordering and supply based on the trust relationship can contribute to an apparent reduction in the variability of inventory levels. This result can be explained by the fact that mutual learning and trust development based on past experiences of trading diminishes an agent’s uncertainty about the trustworthiness of his trading partners and thereby tends to stabilize inventory levels.
ABS is also employed to study the supply chain risks and uncertainty. Supply chain risks can refer to those unexpected and disruptive events that can cause instability and increased costs in operations. Due to the numerous interacting factors that contribute to increased vulnerability and uncertainty in supply chains, traditional methods might be inadequate for the management of supply chain risks and uncertainty. ABS represents a recent development in supply chain modelling which can address the dynamic behaviours of risk/uncertainty issues and which has been regarded as highly appropriate for studying risk/uncertainty management.
The ABS modelling of supply chain risks can be carried out in three different but not mutually exclusive aspects, namely the identification and creation of various types of risk events, risk management, and performance measures of the risk impact and coping strategies. For instance, Sirivunnabood and Kumara (2009) modelled a supply chain network under supplier risks, in which four types of risks were imitated,including rare and short, rare but long, frequent but short, and frequent and long risks. In addition, two risk mitigation strategies (having a redundant supplier and reserving more inventories) were applied to compare the performance. The results highlighted that both approaches are effective but are subject to different risk conditions. In another work, Ehlen et al. (2014) studied how a particular chemical supply chain could potentially behave during and after disruptive events, and how the operation of the supply chain could be affected by the disruptions in terms of scope and duration.The results of this model were used to inform homeland security policymakers about how to prepare better for, prevent, and mitigate losses to the U.S. chemical sector.
Most ABS work to study the supply chain uncertainty has focused on the demand uncertainty and have proposed various approaches for improving the management of the uncertainty, such as the model described in Datta and Christopher (2011). They modelled the effectiveness of several proposed mechanisms for reducing the demand uncertainty in a make-to-stock supply chain. The results indicated that a centralised information structure without widespread distribution of information and coordination is not effective in managing the uncertainty within supply chain networks. In another example, Hing Kai and Chan (2006) proposed a mechanism with an early order completion contract in order to improve the supply chain costs and the order fill rate under the impact of the demand uncertainty. Various levels of the demand errors and variations were modelled throughout the modelling process to evaluate the effects of the proposed mechanisms on supply chain performance.
ABS is often used to study the supplier selection problem in a supply chain,given that procurement is one the three main functions of SCM (the other two are manufacturing and logistics). Procurement is crucially important to SCM because about 50—70% of the costs of a final product are paid to suppliers, which means that those suppliers are responsible for more than half of the overall value-added activities in a typical supply chain. It is therefore worthwhile to make careful decisions when selecting suppliers to guarantee both the quality and quantity required for the supplier activities. ABS can be one effective decision support tool to model the various scenarios and mechanisms for supplier selection, given that it can capture the micro complexity associated with one unique agent’s learning and evaluation process.
Yu and Wong (2015) built a negotiation model to evaluate a supplier selection process that involves a bundle of products with synergy effects. They showed that the purchasing company and suppliers can reach agreements on the details of products simultaneously and thus exploit the synergy effect between products.
Fu-ren et al. (2005) studied trust as a criterion for supplier selection in a complex three-tier supply network. Within their model, manufacturers can select suppliers based on the perceived degree of trust in their partners and on their current quotations. The ratio of these two factors to one another modulates the supplier selection decision, which may also affect the subsequent supply chain performance. The results found that the proposed trust mechanism helped to reduce the average cycle time and increase the in-time order-fulfilment rate in certain market environments but at the expense of increased material costs. Furthermore, a higher trust or propensity to trust leads to a higher in-time order-fulfilment rate. Conducting supplier selection using the trust mechanism is therefore better than using the only quote price and due date.
Liu et al. (2014) presented a multi-criteria decision-making approach to support the selection of appropriate suppliers. Two important evaluation elements for supplier selection related to trust and reputation were considered and a decision model for supplier selection was developed based on these elements to evaluate the performance. The simulation experiments demonstrated that the proposed trust and reputation model can effectively filter unfair rating scores to evaluate the trustworthiness of suppliers. In addition, due to the proposed multi-criteria decision making method, customers would select the most suitable supplier rather than the best supplier in the supply chain.
Schieritz and Grobler (2003) studied how the different order fulfilment strategies and supplier attractiveness affect supplier selection and the supply chain structure.Downstream agents rely on a performance evaluation mechanism based on system dynamics to measure and record the supplier’s delivery performance, and upstream agents adopt either a FIFO strategy or a relationship-based strategy to fulfil orders.The results showed that when using the FIFO strategy, every possible link between the customer and supplier agents is realized and suppliers are switched frequently.On the contrary, the relationship-based strategy supports the development of fixed preferences that lead to a long-term relationship between a customer and his supplier and therefore to less supplier switches. Furthermore, if the customer places more value on the past performance of suppliers, significantly fewer supplier switches and a more stable supply chain structure can be observed, even when suppliers fulfil orders using the FIFO strategy.
These studies show that the main advantage of developing models concerning the supplier selection problem using ABS is mainly because ABS can capture the internal state of mind of an agent to reflect its evaluations of the supplier’s performance and adaptions to the choice of suppliers over time.