Computers and Electronics in Agriculture 76 (2011) 69–79 Contents lists available at ScienceDirect Computers and Electronics in Agriculture journal homepage: www.elsevier.com/locate/compag A novel decomposition and distributed computing approach for the solution of large scale optimization models Yogendra Shastri ∗ , Alan Hansen, Luis Rodríguez, K.C. Ting Energy Biosciences Institute &, Department of Agricultural and Biological Engineering, University of Illinois at Urbana-Champaign, 1304 W. Pennsylvania Avenue, Urbana, IL 61801, United States article info Article history: Received 3 August 2010 Received in revised form 22 December 2010 Accepted 15 January 2011 Biomass feedstock Optimization Computation Agent-based modeling Decomposition abstract Biomass feedstock production is an important component of the biomass based energy sector. Seasonal and distributed collection of low energy density material creates unique challenges, and optimization of the complete value chain is critical for cost-competitiveness. BioFeed is a mixed integer linear pro- gramming (MILP) problem model that has been developed and successfully applied to optimize bioenergy feedstock production system. It integrates the individual farm design and operating decisions with trans- portation logistics to analyze them as a single system. However, this integration leads to a model that is computationally demanding, leading to large simulation times for simplified case studies. Given these challenges, and in wake of the future model extensions, this work proposes a new computational approach that reduces computational demand, maintains result accuracy, provides modeling flexibility and enables future model enhancements. The new approach, named the Decomposition and Distributed Computing (DDC) approach, first decomposes the model into two separate optimization sub-problems: a production problem, focusing on on-farm activities such as harvesting, and a provision problem, incorporating the post-production activities such as transportation logistics. An iterative scheme based on the concepts from agent based modeling is adapted to solve the production and provision problems iteratively until convergence had been achieved. The computational features of the approach are further enhanced by enabling distributed computing of the individual farm optimization models. Simulation studies compar- ing the performance of the DDC approach with the rigorous MILP solution approach illustrated an order of magnitude reduction in computational time using the proposed DDC approach. Moreover, the solu- tion obtained using the DDC approach was within ±5% of the rigorous MILP solution. This approach can be a valuable tool to solve complex supply chain optimization problems in other sectors where similar challenges are encountered. © 2011 Elsevier B.V. All rights reserved. 1. Introduction The importance of biomass feedstock production and provision in the success of the biomass based energy sector has been increas- ingly highlighted in recent times (Perlack et al., 2005; DOE, 2008). Low energy density, seasonal availability and distributed supply create unique challenges that need to be addressed effectively. Novel feedstock alternatives as well as new technologies for crop management, harvesting and post-harvest storage and handling are being developed and made commercially available. Moreover, since a number of independent farms supply feedstock to a single refin- ery, farm level decisions have direct implications on transportation ∗ Corresponding author at: Energy Biosciences Institute & Department of Agricul- tural and Biological Engineering, 1206 W. Gregory Drive, Urbana, IL 61801, United States. Tel.: +1 217 333 1775; fax: +1 217 244 3637. E-mail address: yshast1@illinois.edu (Y. Shastri). logistics and refinery operations. This makes the selection of the optimal sequence of operations critical and non-trivial. Shastri et al. (2009) emphasized the importance of taking a sys- tems approach to overcome these challenges and achieve a system level optimal configuration that ensures seamless integration of various production tasks. BioFeed is an optimization model which has been developed as a first step towards such a systems analysis framework (Shastri et al., 2009, 2011). It is a mixed integer lin- ear programming (MILP) model that incorporates various biomass production activities such as harvesting, packing, transportation, storage and handling, and determines the optimal system level configuration. An important unique feature of the model is the optimization of the operational blueprint for the whole system that can be used by farmers and managers. The model has been successfully applied to the case of switchgrass and Miscanthus production in southern Illinois (Shastri et al., 2009, 2011). How- ever, the model simulation and optimization studies highlighted its computational complexity. This is primarily due to the large 0168-1699/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.compag.2011.01.006