Energy represents the largest managed operating cost for the hydrocarbon processing industries. The best opportunities for improving plant profitability lie in the examination of the design and operations of the site’s utility system. KBC’s Linnhoff March Energy Services specializes in energy improvement and has over 10 years experience in utilities modeling and analysis. We offer a variety of services, including our advanced OptiSteam package. This provides utility management and optimization tools that automatically determine least-cost operational utility configurations, while taking into account fluctuations in process demand, energy costs, and environmental restrictions.
A good example of such a system was developed for Cepsa’s Gibraltar refinery. This 220,000 bbl/d refinery is the largest of three owned by Cepsa (Compañía Española de Petróleos, SA). This complex site significantly reduced its utility costs through the installation of two 37MW gas turbines equipped with heat recovery steam generators. While this project was a success, the new configuration resulted in a step-change in the complexity of the system. In addition, intricate contractual arrangements for power and fuel were imposed by the Spanish legislature. These combined factors made it very difficult to operate the utility system in a truly optimal way and highlighted the need for a tool to assist Cepsa’s engineers and operators in this task. After a competitive bidding process, Linnhoff March was selected to develop an OptiSteam-based optimizer model of the refinery utility system, capable of handling the complexities of the physical system and the intricacies of the utility contracts.
The Figure illustrates the structure of the system implemented at the Gibraltar refinery. The system can be imagined as a set of building blocks, which can be assembled together to form a powerful, coherent structure. The foundation of the structure is the detailed model of the utility system, including all the unit operations and associated thermodynamics. This core model is then used to perform a wide range of tasks.
Referring to the “Off-line” section of the Figure, the user has full control of the inputs allowing the model to perform simulated “what-if” analysis. The economic impact of refinery changes can be evaluated by considering power import and export of fuels and water at the battery limit. In fact, this is the ideal environment to put all project ideas that might be fighting for the same capital funding into a common perspective, since all the constraints and interactions are taken into account to define the true operating costs. An example of this is the use of the model as a support tool in the evaluation of a new cogeneration plant, which is currently in the construction phase. This plant, partly owned by Cepsa, will sell steam to the refinery and will greatly affect the optimization of the utility system.
Regarding off-line optimization, the model must find the least-cost operation that meets the process demands, restricted by physical constraints and equipment availability. The optimization problem is complex and has both continuous and integer variables with some strong non-linearities. To solve this, it was necessary to use state-of-the-art solver engines and apply a combination of Mixed Integer Linear Programming (MILP) and Non Linear Programming (NLP) techniques.
The main role of the optimizer when run off-line is that of a planning and forecasting tool. In Cepsa’s case, the model had the ability to forecast utility demands for the refinery processes since they developed correlations of unit throughput and demand. With that, the tool was able to analyze production plans over any selected dates. Typically, depending on the dates selected for analysis, it becomes necessary to model (and optimize) several discrete operating scenarios. The system interface guides the user through this complex task and quickly produces Excel-based reports showing the optimized results from each scenario, plus a summary of the optimized plan.
The “On-line” section of the Figure shows the software using current plant data from a
Honeywell Plant Historian Data (PHD) system to perform real-time optimization. Before
the raw data can be confidently used, it undergoes preliminary validation and reconciliation steps in dedicated routines developed by Linnhoff March. First, it is validated against predefined bounds and user overrides before statistical noise filtering is applied. The pre-filtered data is then sent forward to the data reconciliation model. Larger problems with more than 900 constraints and 1,500 variables are solved using a powerful Quadratic Programming (QP) engine. The reconciled data is reported along with the results from statistical tests, giving an indication of the confidence of each measured value. This information can be used to identify faulty or inaccurate meters and consequently increase the precision of the plant data.
Once a reconciled set of real-time data is produced, it is possible to determine the optimum operational configuration. The main degrees of freedom available are: boiler selection and loads, turbogenerator operation and extraction/condensing flows, fuel selection, and gas turbines operation (power generation, steam injection, etc.). Unlike the off-line model, which can run at every engineer’s desk, the on-line model is hosted in a dedicated, powerful Windows® 2000 server. With no direct user interaction at all, the data collection and filtering runs every 10 minutes, while the reconciliation and optimization runs every 30 minutes. There are three different levels of user access to allow for changes in equipment availability (control room level), prices and tariffs (managers and engineers level) and model structure (programmer level).
As the reconciliation and optimization is completed, the system automatically exports the results in two ways: web-based reports (a current set of results is published to a set of web pages on the company intranet) and a custom-built SQL Server database (Cepsa’s preferred choice). Other options are storing results in the site data historian or an Access® database. In each case, status flags reporting the feasibility of the solution, the “health” of the software, and other key items are included with the results.
The system described in this paper was developed during the first half of 2003 and it was implemented at the Gibraltar refinery during the summer. The latest results available at the time of publication were analyzed to provide an indication of the savings predicted by the model resulting in more than 350 completed optimization runs showed an average reduction in operating cost of 3%.
In summary, the model-centered system described can be used to simulate different “what if” scenarios to validate and reconcile data, or to optimize the operation by finding the least-cost configuration. While it can run off-line (manual execution), its most sophisticated mode is linked to the refinery data historian and works as an on-line, openloop optimization tool. This user-friendly system presents the results by generating web-enabled reports and stores the main variables in an SQL Server database.
