An optimization-based web application for synthesis and analysis of biomass-to-fuel strategies

: We develop an optimization-based web application for assessing biomass-to-fuels strategies based on the Biomass Utilization Superstructure (BUS) framework. This web application allows researchers with limited knowledge of optimization to assess different technologies employed at different strategies and identify the major cost drivers. The user must only provide the necessary parameters and create an optimization run after identifying the question to be addressed and the assessment metric. The web application generates visual representations of the results once the optimal solution is obtained. Finally, we demonstrate the applicability of the web application through an example. © 2017 The Authors. Biofuels, Bioproducts and Bioreﬁ ning published by Society of Chemical Industry and John Wiley & Sons, Ltd. Supporting information may be found in the online version of this article.


B
iomass is a renewable energy source that can potentially mitigate global climate change linked to greenhouse gas (GHG) emissions. 1,2 Advances in fundamental research have resulted in the development of numerous biomass-to-fuels strategies. Th e abundant biomass, including energy crops 3 and biomass wastes, 4,5 can be deconstructed and converted to biofuels biochemically, 6-8 catalytically, 9,10 or thermo-chemically. 11,12 Evaluation of these strategies is key in identifying major cost drivers and bottlenecks for biofuel production. Th e major challenge in this type of analysis is that feedstocks can be converted to biofuels in multiple ways and, in general, a biofuel can be produced by multiple feedstocks. An eff ective way to evaluate multiple options simultaneously is superstructure optimization, [13][14][15][16] which is a modelbased approach that considers alternative strategies simultaneously. Based on this approach, Kim et al. developed a Biomass Utilization Superstructure (BUS) framework, 17 which incorporates a wide range of conversion technologies along with the corresponding feedstocks, intermediates, and fi nal products. Th is framework allows the identifi cation of new biofuel production strategies, and the simultaneous assessment and comparison of alternative strategies.
Nonetheless, generating a superstructure, and developing and solving the resulting optimization model are complicated and can be especially challenging for researchers without domain expertise. Th erefore, to facilitate the implementation of the BUS framework, which allows researchers across diff erent fi elds to generate key insights into the use of newly developed technologies, we develop a web application for assessing biomass-to-fuels strategies, which is accessible to the public through https://bus.glbrc.org/. Th is web application off ers the user the fl exibility to select diff erent products or feedstocks of interest and assessment metrics (e.g. cost minimization or profi t maximization).
In the remaining text, we fi rst present the problem statement and model formulation. Th en we describe the diff erent types of questions that can be addressed using this web application and its basic structure. Finally, we demonstrate how it can be used to address a range of problems.

Optimization problem
Problem statement Figure 1 shows a biomass utilization superstructure which consists of potential deconstruction and conversion technologies, and the corresponding compounds. We treat technologies and compounds as nodes, and compound fl ows as arcs. We classify compounds into four categories: feedstocks, intermediates, products, and by-products. A strategy consists of a subset of technologies and the associated compounds, and it is obtained as a solution of the superstructure model. For example, in a biochemical conversion strategy, corn stover is pre-treated and converted to ethanol via dilute acid pre-treatment, enzymatic hydrolysis, and co-fermentation, followed by product purifi cation. 18 We use lowercase Greek letters for parameters, uppercase italic Latin letters for variables, lowercase italic Latin letters for indices, and uppercase bold Latin letters for sets.
Th e general problem is stated as follows: We are given a set of compounds i ∈ I with a vailability a i , mi nimum purchase b i , minimum and maximum demands γ i γ i , and price l i . Th e subsets of feedstock, intermediate, product and by-product are denoted by i ∈ I F /I I /I P I B , respectively. We are also given a set of technologies j ∈ J with capacity d j , unit production cost m j , and conversion coeffi cient h i,j . Th e subsets of technologies that produce and consume compounds are denoted by j ∈ J i / j i + -, respectively. We aim to identify an optimal strategy to meet feedstock consumption or fi nal product demand targets using various metrics.

Mathematica l formulation
We introduce the following non-negative continuous variables: P i denotes the amount of feedstock i ∈ I F , purchased; S i denotes the amount of product or by-product i ∈ I P ∪ I B , sold; X j denotes the production level of technology j.
Th e material balance is given as: Note that h i,j <0 for inputs and h i,j >0 for outputs. For each technology, we fi x the conversion coeffi cient for the main product to 1, and normalize the coeffi cients for the remaining components that enter or exit the technology (Supporting Information).
Th e feedstock consumption is bounded by its minimum purchase and availability: Th e production level of each technology is bounded by its allowable capacity.
Th e sales of products and by-products are bounded by their minimum and maximum demands. Th e evaluation can be performed using various metrics. For example, we can use minimum cost, which includes the feedstock purchase and the production costs, minus the sales of by-products.
We can also use maximum profi t as an objective function, which includes the sales of products and by-products, minus the feedstock purchase and the production costs.
Th e linear programming (LP) model is coded in General Algebraic Modeling System (GAMS) 19 and solved using CPLEX solver.

Types of questions
We can use the proposed model to address diff erent types of questions, such as: Question 1: What is the optimal strategy for a specifi c feedstock?
Question 2: What is the optimal strategy for a specifi c product?
Question 3: What is the optimal strategy for specifi c feedstock and product?
Th e details on how the models must be modifi ed to address these questions are given in the Supporting Information. Once a question is chosen, we formulate a problem by selecting the metric (e.g. cost minimization or profi t maximization) to be used to fi nd the optimal strategy. Th e problem is represented as Q.Z, where Q is the question and Z is the assessment metric (Fig. 2). For example, we defi ne problem Q1.Z2 if we aim to fi nd the optimal strategy for a given feedstock (Q1) using profi t maximization as the metric (Z2).

Web application interface
Th e BUS web application is built on the Ruby on Rails 20 web application framework and hosted at the Great Lake Bioenergy Researcher Center (GLBRC). To begin using the application, the user must fi rst create an account. Figure 3 depicts the general workfl ow within the web application. Th e user is only required to enter the necessary parameters in the Network page, and identify the question to be addressed and the assessment metric in the Optimization page. If there are any errors, warnings will appear to alert the user. Th e details on Network and Optimization pages will be discussed in the following sub-sections. All stored parameters will be exported to the optimization tool and the LP model will be solved. Th e application then stores optimization results and generates visual representations. Th e user can view and clone the existing network or optimization results as well as share it with other users.

Network
Th ere are three main elements in the Network page: Compounds, Technologies, and Links. Th e user fi rst defi nes the technologies of interest and their corresponding compounds and then provides the necessary parameters (e.g. price, availability, demand). It is then followed by inserting  conversion coeffi cients to connect the compounds and the technologies. Th ere are two methods to insert compounds and technologies within a network: (i) manual entry and (ii) spreadsheet upload. Th e former can be implemented by creating compounds and technologies and fi lling out the data manually. If the network involves large numbers of compounds and technologies, users are encouraged to use the latter method. A spreadsheet template is readily downloadable for users to fi ll in all data. Th e server will automatically process the uploaded data in the database. Finally, the user will be able to see a visual representation of the network that connects all technologies and compounds.

Optimization
Following the creation of a network, the user proceeds to the Optimization page to create a new optimization run. Th e user selects the question to be addressed and the compounds of interest, as well as the assessment metric. Th e optimal strategy is found once the optimization model is solved. Th e Optimal Strategy tab shows the selected technologies and compounds. Th e arcs vary in width based on the fl ows of compounds, and the technology nodes vary in shades of red based on their production costs. A darker color indicates higher production cost while thicker arcs indicate larger compound fl ow. Th e user can see the numerical values (e.g. production cost, compound fl ow, etc.) by hovering over the nodes and arcs. Th e Results tab shows the detailed results of the optimal strategy (e.g. production level of selected technologies, amounts of compounds sold and consumed, cost contributions) in bar charts.

Application Example
We consider the utilization of corn stover and hardwood to produce gasoline, diesel, and ethanol. Figure 4 shows the corresponding simplifi ed superstructure, where we group similar technologies or compound into one node. For example, there are two diff erent dilute acid pre-treatment technologies that consume corn stover and hardwood, respectively. We group both technologies into one node, which denoted as 'DA.' Th e full superstructure is given in the Supporting Information. In thermochemicalbased conversion strategies, corn stover and hardwood are chopped prior to pyrolysis or gasifi cation. We consider two pyrolysis technologies 21 (in situ and ex situ upgrading of fast pyrolysis vapors) and two gasifi cation technologies 22 (indirect and direct gasifi cation). Crude bio-oil and char are produced from pyrolysis. Th e former is then converted to gasoline and diesel via hydrotreating and hydrocracking, while the latter is sent to combined heat and power plant for electricity generation. Th e raw syngas produced from gasifi cation is sent for steam reforming, followed by methanol synthesis. Methanol is either sent to acetic acid synthesis and hydrogenation for ethanol production. Acetic acid can be sold as a by-product. In biochemical-based conversion strategies, corn stover and hardwood can be pre-treated using dilute acid 18,23 or ammonia fi ber expansion 6,17 pre-treatments. Th e hydrolyzate is either sent to separate acidic or enzymatic hydrolysis and followed by fermentation; or simultaneous saccharifi cation and fermentation. Th e culture broth is then separated using distillation to obtain high purity ethanol. Th e remaining residue can be utilized in combined heat and power plant to generate electricity, which can be sold to the market. We assume that the prices 6,17,18,20-22 of corn stover, hardwood, acetic acid, and electricity are 0.074 $ dry kg -1 , 0.074 $ dry kg -1 , 0.789 $ kg -1 , and $0.059 $ kWh -1 , respectively. All technical and economic parameters are provided in the Supporting Information. All costs are indexed to 2016 dollars. Note that the consumption of utilities and auxiliary inputs, (e.g. make-up water, solvents, catalysts) are included in the calculation of costs but they do not appear as compounds in the superstructure.
Next, we solve the following problems using the web application: • Q1.Z2: most profi table utilization strategy for corn stover. • Q2.Z1: most cost-eff ective production strategy for ethanol. • Q3.Z1: most cost-eff ective production strategy for hardwood-to-gasoline.
We show the results and discuss the key cost drivers in the next sub-section.

Results and discussions
In Problem Q1.Z2, we assume the market price of ethanol to be 0.67 $ kg -1 (3 $ GGE -1 ). Th e maximum profi t is -0.04 $ kg -1 corn stover consumed. Th e negative profi t value means that the revenue from ethanol (0.26 kg kg -1 corn stover) and electricity sales (0.16 kWh kg -1 corn stover) are lower than the sum of feedstock and production costs. Th e optimal strategy includes dilute acid pre-treatment, enzymatic hydrolysis, fermentation, distillation and waste-water treatment, and combined heat and power (Fig. 5(a)). Note that the fl ow of hydrolyzate is larger than the fl ow of corn stover due to water, which is not defi ned as a compound here, is added in the pre-treatment. We calculate the breakeven price to cover the cost of this strategy, which includes feedstock purchase and production costs, minus the sales of by-products. Th e breakeven price is 3.78 $ per gallon gasoline equivalent (GGE). Th e main cost driver is feedstock cost (34% of breakeven cost) ( Fig. 6(a)). In terms of production costs, distillation, and waste-water treatment (20% of breakeven cost) is the major cost contributor and it is followed by enzymatic hydrolysis (17% of breakeven cost).
In Problem Q2.Z1, the minimum cost is 0.80 $ kg -1 ethanol (minimum selling price, MSP = 3.64 $ GGE -1 ). Th is strategy shows 2.86 kg of hardwood is consumed to produce 1 kg ethanol, and 0.33 kg acetic acid is produced as by-product. Th e selected technologies are handling and chopping, direct gasifi cation, steam reforming, methanol synthesis, acetic acid synthesis, and hydrogenation ( Fig.  5(b)). Hydrogenation (1.50 $ GGE -1 ) and acetic acid synthesis (1.45 $ GGE -1 ) are the major cost drivers, accounting for 41% and 40% of MSP, respectively ( Fig. 6(b)). Th is user-friendly application allows researchers with no background on optimization to easily assess strategies employing technologies of interest. Th e user only needs to insert the technologies and compounds data since the application integrates data processing, optimization, and visual representation. Furthermore, this application allows the user to save the inserted data and the optimization results, and share them with other users. Beyond the user-friendly interface, the application has the fl exibility to address different questions using various metrics. We hope this web application will help researchers in identifying economic drivers of their strategies, thereby guiding their future research eff orts.
is due to the high cost of make-up carbon monoxide and hydrogen in this strategy. 22 In Problem Q3.Z1, the minimum cost is 1.54 $ kg -1 gasoline produced, which yields a MSP of 3.24 $ GGE -1 . In this strategy, 1 kg gasoline and 0.37 kg diesel are produced when 5.77 kg of hardwood is utilized. Handling and chopping, pyrolysis with in situ upgrading, hydrotreating, hydrocracking, combined heat and power are selected (Fig. 5(c)). Th e residue from the pyrolysis reactor is combusted and the excess electricity (0.22 kWh kg -1 gasoline) can be sold to the grid. Th e major cost contributor is pyrolysis (48% of MSP), due to high capital cost of reactors 21 (Fig. 6(c)).

Concluding remarks
We developed a web application based on the BUS framework for assessing biomass-to-fuels strategies. Th is