An expert system design for a crude oil distillation column with the neural networks model and the process optimization using genetic algorithm framework

https://doi.org/10.1016/j.eswa.2007.08.105Get rights and content

Abstract

In this study an expert system of a crude oil distillation column is designed to predict the unknown values of required product flow and temperature in required input feed characteristics. The system is also capable to optimize the distillation process with minimizing the model output error and maximizing the required oil production rate with respect to control parameter values. The designed expert system uses the practical data of an operating refinery located in Abadan. The input operating variables of the column were operational parameters of crude oil such as flow and temperature, while the system output variables were defined as oil product qualities. We can make the knowledge database of these input–output values of plant with the aid of a neural networks model (NNM) to organize and collect all data related to this process and also to predict the unknown output values of required inputs. In addition we have made the ability of system’s optimization with the use of genetic algorithm (GA) with the aim of error minimizing of expert system’s output and also maximizing the required product rate with respect to its industrial importance. The built expert system can be used by operators and engineers to calculate and get some unknown data for operational values of this distillation column.

Introduction

An expert system is a computer system employing expert knowledge to attain high levels of performance in solving the problems within a specific domain area. By encapsulating expert knowledge and experience, expert system enables organizations to support important decision-making and improve organization productivity (McCarthy, 1984). Today, expert systems have demonstrated their potential, gained credibility, and are being widely used to solve a variety of problems in industry and government. Generally, the basic structure of an expert system consisted of a knowledge database and an inference engine to reason a proper answer to the domain user. The knowledge database, a core of the system structure, was built using the collection and organization of the experts’ experiences in a particularly defined system (Rich, 1994). Methods such as fuzzy logic, artificial neural networks (ANN), or neuro-fuzzy, are generally used to construct the knowledge database (Lababidi & Christopher, 2003). This database can be represented by a set of rules, a pattern, or even topological figures, relating the input to the output of the operating system (Jackson, 1999). One of these complex problems for the control of which an expert system is amenable, is a crude oil distillation column. Distillation column control is one of the important problems of production/operations management area, as small improvements in the performance of the system can lead to significant monetary consequences. It is of utmost importance to develop practical solution procedures which yield high-quality design decisions with minimal computational requirements. Crude oil distillation is used to separate the hydrocarbons in crude oil into fractions based on their boiling points. The separation is done in a large tower that is operated at atmospheric pressure. The tower contains a number of trays where hydrocarbon gases and liquids interact. The liquids flow down the tower and the gases up. The lighter materials such as butane and naphtha are removed in the upper section of the tower and the heavier materials such as distillate and residual fuel oil are withdrawn from the lower section. A distillation column is generally used to separate mixtures in petrochemical industries and so a crude oil distillation column is one of the most important parts in refining industry. The practical goal of such a kind of unit is to achieve a higher production rate beside the lower costs relate to economic consideration; so in a crude distillation process, the objective is to perform a process optimization including high production rate with required product quality and low operating costs by searching an optimal operating condition of the operating variables (Sea, Oh, & Lee, 2000). The crude separation process involves many complex phenomena which have to be controlled in its best placement. The input variables of crude distillation column are usually energy supply inputs, reflux ratios, and product flow rates, while the output variables are the oil product qualities, system operating performance, or the plant profit. However, because of the non-linear interactions between the operating input and output variables, search and maintain in an optimal operating condition is a complicated job. In addition, the optimal manipulated variables of CDU have to be frequently adjusted due to the variation of crude oil properties. Furthermore, if specifications of oil products cannot be reached, the oil supply can cause some problems in plant management. All these sequences raise the need to control and optimize the complex crude operation. During the past two decades, there has been a growing awareness among academia and industrial practitioners that operability issues need to be considered explicitly at an early phase of process design. Mathematical frameworks have been developed for incorporating steady-state flexibility requirements into process design (Grossmann & Straub, 1991). In recent years, the research of crude distillation process was focused on the subject of process control and optimization (Mizoguchi, Martin, & Hrymak, 1995). Complex processes generally inherit higher non-linearity and uncertainty of the system models. In a petroleum plant, optimal variables of a distillation unit process were decided mostly by experienced operators according to the input crude oil conditions. However, if the experienced operators are not available all the time for a continuous operation or the decision-making is not correct, the product quality specification or even the optimal operation cannot be reached. Therefore, available support for the optimal operating information to the operators is quite essential to maintain a proper management of the crude operation. Expert systems, one field of Artificial Intelligence, apply expertise to provide solutions for many complex systems in recent years (Giarratano & Riley, 1993). Also ANN approach has been found as one of the effective ways to model complex processes due to the non-linear characteristic of the ANN structure. The objective of this work is to implement the constructed expert system for providing optimal operating information in defined optimization problem and predict the output quality of the crude distillation column with respect to defined input variable considerations. The scheme to construct the expert system is to first build the knowledge database of the distillation column operating system. The database was built using a neural networks approach. This knowledge database, represented and stored by an ANN configuration, was then used to estimate the optimal operating conditions of the process using the genetic algorithm optimization method.

Section snippets

Refinery operations

Refineries are composed of many different operating units that are used to separate fractions, improve the quality of the fractions and increase the production of higher-valued products like gasoline, jet fuel, diesel oil and home heating oil. The function of a refinery is to separate the crude oil into many kinds of petroleum products (Kary & Leif, 1993). In this project, the input variables of process are the crude oil properties, such as its flow rate and the temperature as control variable,

Research design and method

The design of the distillation column’s expert system for process optimization was done in three steps.

The first step was to establish the model database of the operation from a series of practical data. The analytical data were measured and collected in a real petroleum plant. In the second step, the plant model has constructed by the neural networks modeling approach using the input–output data of the column. Total 180 sets of the input–output data were used in the ANN modeling for the crude

ANN modeling

As described in the previous sections the neural networks model of the practical distillation column is constructed to represent the database of the system. This model uses the operational data of the refinery and is constructed in two steps of train and test. As a first step, we fed 60% of data to the net and train it with, and then use the rest 40% randomly of data to test the net. The result of comparison between the real output of plant and the model’s answer to the test is illustrated in

Conclusions

The expert system of the crude oil distillation was found to predict the optimal operating conditions for the objective function. The knowledge database was well established using ANN model with expertise of the unit operators. The neural networks model can represent and describe the distillation process for the input (system operation) and output (product quality) relations. Practical optimization cases of minimizing the cost function of oil production rates were estimated by solving the

References (10)

  • H.M.S. Lababidi et al.

    Web-based expert system for food dryer selection

    Computers and Chemical Engineering

    (2003)
  • J. Giarratano et al.

    Expert systems: principal and programming

    (1993)
  • Grossmann, I. E., & Straub, D. A. (1991). Recent developments in the evaluation and optimization of flexible chemical...
  • S. Heykin

    Neural networks

    (1999)
  • P. Jackson

    Introduction to expert systems

    (1999)
There are more references available in the full text version of this article.

Cited by (73)

  • Refining data-driven soft sensor modeling framework with variable time reconstruction

    2020, Journal of Process Control
    Citation Excerpt :

    As the rapid development of distributed control system (DCS) and industrial cloud computing, process industries are entering the big data era [1,2]. Data-driven models have been widely used in monitoring [3], control [4] and optimization [5]. As a typical application, data-driven soft sensors have been substantially implemented in process industries [6], which have greatly improved the modeling efficiency.

View all citing articles on Scopus
View full text