Apr-2025

Enterprise-wide optimisation across hydrocarbon processing assets

Consider the layers of efficiency when establishing a global resource optimisation modelling system.

Jun Yi, Pei Su and Yonglei Wang, Galaxy Sky-grand Technology Co., Ltd.
Weijun Yang, PetroChina
Jinxiang Mao, Sinopec
Ricky Hsu, Independent Consultant

Article Summary

Enterprise optimisation problems can be multi- plant, multi-period, or a combination of both, as well as pipeline or production plans and schedules. The Global Resource Optimization Modeling System (GROMS), used to optimise large multi-refinery enterprise problems, has been shown to increase profit potential by $ per ton ($0.25/barrel) of crude oil feed vs other commercially available programs.

Advanced characteristics
GROMS uses a database platform with a basic structure defined by the ‘model business dictionary’. Users only need to define and manage a ‘business dictionary’. There is no need to define and manage ‘code dictionaries’ once defined for periods, companies, production equipment, processing cases, logistics, physical properties, and recursive variables.

The GROMS math matrix is generated entirely by the mixed-integer nonlinear programming (MINLP) algorithm dictionary and can simultaneously detect the correctness of the data entered by the user.

Model builders and users are focused on the business and do not need to clearly define all model structures for purchasing, primary and secondary processing, product mixing, and sales. There is also no need to be concerned about the relationship between these structures.

The GROMS model structure was originally developed as a mathematical matrix for ‘multi-enterprise, multi-cycle, multi-business, and multi-goal’ business relationships. The MINLP algorithm, implemented in C++ language, directly operates the model matrix (in matrix product state [MPS] format) and automatically generates the initial value of the coefficients required for solving with a commercially available solver. The model system has been designed according to this architecture and systematically optimised for performance on 10 years of model experience. Larger problems can, therefore, be solved faster with a low occurrence of local optimum.

User interfaces
The programming efficiently utilises a database platform with a unique and easy-to-use user interface to build models, manage case data input, and analyse results with user-customised reports.
The GROMS main interface (see Figure 1) includes:
• Top menu: Selection area of ‘function menu, model filter conditions’.
• Model tree (left column of Figure 1): Display and maintain ‘model classification, logistics structure, logistics relationship’, and more.
• Main data window: Secondary unit yields and controls, which generally display and maintain all constraint data, such as model logistics, integers, prices, physical properties, operating conditions, and pipelines.

The calculation interface (see Figure 2) includes:
• Window 1: Select the model to be calculated.
• Window 2: Select the CASE to be calculated and display the calculated target value and convergence status.
• Window 3: Select the model ‘logistics, physical properties’ constraint group. Constraint groups that are not selected are not included in the calculation, resulting in convenient step-by-step commissioning.
• Window 4: Set the parameters of the MIN3LP algorithm. When there is a computational problem or a local optimal solution, it can usually be solved by adjusting the calculation parameters.
• Middle toolbar: Function buttons such as calculation operations, result viewing, and report charts. GROMS quickly customises the output of the process flow diagram.
• Lower window: Displays the calculation log information of the MIN3LP algorithm.
The result browsing interface (see Figure 3) includes:
• Select row: Select model, CASE, and comparison CASE to be browsed.
• Toolbar: Filter information and operation buttons required for browsing functions.
• Data window: Information on constraints and results for both the basic CASE and the comparison CASE.

Model types
The modelling options or types available are:
• LP: Linear programming
• MILP: Mixed integer linear programming

Mixed integer (MIP) constraint types can be directly defined, and MIP constraint data can be entered, such as ‘start-stop sign’, ‘batch and the number of batches’, ‘threshold’, ‘sequence’, ‘indicator light’, ‘route selection’, ‘pass and no pass’, ‘stop’. All of these are expressed by business methods, and there is no need to write MIP mathematical equations manually.

Business model and mathematical matrix construction technology
GROMS includes functions such as model building and management, matrix generation and solution, model debugging and result browsing. It also offers customised report/chart output, configuration tools and system management. By adopting advanced technology, it offers a general planning optimisation modelling system based on business rules, with the following characteristics:
• Business-based modelling: Users can establish models according to business relationships, making it easier for them to build a new model. The resulting model is the standardised expression of business, so it has good readability. At the same time, there is no need to manually write objective functions, variable codes, and constraint equations. Users can directly establish ‘multi-period, multi-enterprise, multi-business and multi-objective’ models.

The model’s ‘all logistics, weight, volumetric properties, density, and more’ is automatically transferred, and there is no need for manual transfer. GROMS adopts a database storage model, and there is no theoretical limit on the size of the model.
• MIN3LP algorithm technology: The system can automatically transfer physical properties through the entire process and generate recursive initial values for the nonlinear pooling of the flow (with physical properties). The nonlinear distributed recursive algorithm is used to solve the problem, and the recursive convergence accuracy reaches 1.0e-6.

The GROMS-MIN3LP algorithm can usually obtain a global optimal solution when solving the MIN3LP model. It effectively avoids the problem of ‘local optimal solution’ when the MIN2LP algorithm solves the MIN3LP model and addresses the problems associated with the MIN2LP algorithm.
• Matrix generation, calculation, and customise report technology: The core algorithm of GROMS is based on highly readable business models, which can directly and automatically generate standard mathematical matrix MPS files, output the results to the database system after iterative calculation, and provide functions such as synchronous browsing, querying, debugging, and customising reports of models and results.

The user does not need to be associated with a mathematical matrix because the size of the MPS is not limited. Since the actual model size is limited, GROMS sets the upper limit of the MPS matrix rows and columns, both of which are 99999999.

The result output provides details of the entire logistics, physical properties, and transmission process. Users can export the results to Excel and other software. They can also customise functions such as static reports and process flow charts. Models, debugging information, and solution results can be displayed simultaneously.
• Debug step-by-step: The complete refinery model usually includes constraint groups, such as material flow, MIP, density (specific gravity), weight physical properties like sulphur, and volume physical properties like research octane number (RON). It also encompasses operating conditions, mixing and separating in pipelines, as well as conversion between properties and material flows. Additionally, it includes ‘DB structure, physical properties recursion, and piecewise linear’ calculation. GROMS supports step-by-step debugging in a [selected] way. For example, you can debug the [flow only] model by selecting only the [flow constraints] group. You can debug the [flow only + specific gravity] model by selecting the [flow constraints] group and the [specific gravity] constraint. If the [DB structure] parameter is not selected, the DB calculation will not be performed, and so on.