Uncovering System Behaviors in Biofuels Supply Chain Network Using an Agent-Based Simulation Approach

Uncovering System Behaviors in Biofuels Supply Chain Network using an Agent-based Simulation Approach Datu B. Agusdinata

Industrial & Systems Engineering and Environmental and Sustainability & Energy (ESE) Institute, Northern Illinois University

590 Garden Road EB 236, DeKalb, IL, USA

Telephone: (815) 753-6748; Fax. : (815) 753-0823, email:
bagusdinata@niu.edu

The second generation biofuel feedstocks such as camelina, short rotation woody crops, algae, and switchgrass have been investigated to minimize the impact on the food chain. In order to achieve a significant market penetration of the second generation biofuels, some significant challenges must be addressed. These include improving feedstock yield and conversion efficiency. Another critical challenge, which is the focus of this study, is a viable and well-functioning supply chain (SC). To add to these challenges, some undesirable behaviors at a system level have been observed in the corn-based ethanol industry. For instance, as a result of the Energy Independent and Security Act of 2007 mandating ethanol production, refining capacity was added at a faster rate resulting in high ethanol inventories. Increased price of corn has squeezed refineries’ profit margin below the sustainable level. A stabilizing price of oil has made ethanol less economically attractive. For producers, the financial pressure has been exacerbated by dried up capital due to the credit crunch and the recent expiration of a federal subsidy for ethanol blenders. Natural factor plays a role too. Recent drought has caused a spike in corn price leading to temporary shut down or scale back production of many ethanol refineries

An agent-based simulation approach is pursued to understand the dynamics of biofuels SC network. The approach treats each actor along the life cycle stages as an entity with distinct decision rules. The approach can deal better with the fact that some actors do not always act rationally due to bounded rationality. Actors also often use heuristics to generate solutions. It can also factor in the fact that actors learn over time and may change their decision rules as a result. This approach is in contrast to most equation-based models such as system dynamics or utility-based methods, which are based on the assumption that actors are rational and share common interests.

The interests of three supply chain actors are represented: users, biorefineries, and farmers. Each actor type has a binary decision option: adoption or non-adoption of biofuels. This SC network model is characterized by distributed control, time asynchrony, and resource contention among actors and who make decisions based on incomplete knowledge and delayed information. The decision dynamics of these actors are modeled using a computational ecosystem construct. A preliminary set of coupled payoff function for each actor type and each decision is developed to represent interdependencies among SC actors. The SC network behavior is observed in terms of fraction of actors adopting the biofuel option. The SC network shows behaviors ranging from fixed point equilibrium under no delay and perfect knowledge to periodic and chaotic oscillations.

The simulation model was used to evaluate three sets of simple and straightforward subsidy policy. The first policy setup involves a constant subsidy level across the simulation period of 120 months or 10 years. Second type of policy is a subsidy scheme that is evaluated periodically. In the model, every 2 years, the relative payoff between biofuels and non-biofuels options is compared. When biofuels are at disadvantage, the subsidy is implemented. Otherwise, no subsidy is put in place. This policy setting is similar to a situation in which regulators like the U.S. Congress deciding whether or not to continue a certain subsidy scheme. The last type of policy is the subsidy that will be phased out over time. This arrangement is similar to special feed-in tariffs established to incentivize adoption of solar energy.

When Constant subsidy policy is implemented during the 120 months of simulation period, the SC network shows a steady increase in the adoption of biofuel options and reaches equilibrium in about 48 months. As a result of the assumed coupled payoff functions, each actor type arrives at different equilibrium states. At most times past year 4, all 1000 farmers adopt biofuel crops whereas the fraction of refineries dedicating for biofuels (out of 100) fluctuate from about 80% to 100%. The fraction of motorist users hovers at around 42%.

Next, Policy2: Declining subsidy, provide incentives for actors to take biofuel options similar to the effect of Policy1. After around year 6, however, the decreased subsidy level starts to take effect. Biorefineries and users begin to drop biofuel options followed by farmers 2 years later. The attractiveness of biofuel options temporarily picks up again afterwards but eventually drops further at lower level when the subsidy has been completely terminated. Lastly, Policy3: Periodic interval subsidy results in periodic pattern of SC network behavior. For the 10 years period, there is a trend that the peak of each cycle of biofuels adoption increases as time progresses.

A sensitivity analysis reveals that the following parameter impacts on system behavior can be observed:

• Reduced payoff uncertainty lowers the actual number of actors opting for biofuel options. As actors know with more certainty about the payoff, their preference will be slightly downgraded and hence the slightly lower adoption level. However, the pattern of system behavior does not change significantly.

• Reduced re-evaluation rate results in a smoother transition to biofuel options for the three actors.

• Reduced delay time leads to quicker action by actors to take biofuel options as they become favorable.

• Lastly, the combined effect of reduced values of the three parameters results in a quicker and smooth transition and stable equilibrium.

One major implication of these findings is that when supply chain actors have more updated information about the decision of other actors and less uncertain knowledge about the payoff of their decisions, they can reduce the oscillations in the SC network. Overall, the simulation model framework provides a basis for further development including identification and assessment of policies to control undesirable behaviors in a supply chain network. The modeling framework can be adapted and applied to other renewable energy applications such as wind and solar energy.

Keywords: Distributed system, agent-based simulation model, computational ecosystems, biofuels supply chain network