June 2020

Special Focus: Process Optimization

Maximizing diesel hydrotreater profit by multivariable optimization: A case study

A diesel hydrotreater (DHT) is a critical unit within an oil refinery. A DHT processes the diesel range fractions obtained from different units of the refinery—such as from the crude distillation unit and the delayed coker unit—to adhere to diesel fuel market specifications.

A diesel hydrotreater (DHT) is a critical unit within an oil refinery. A DHT processes the diesel range fractions obtained from different units of the refinery—such as from the crude distillation unit and the delayed coker unit—to adhere to diesel fuel market specifications. As the feed to the DHT is often different from the design feed (due to changes in crude oil quality and in operating conditions of the upstream units), real-time data analysis and optimization are necessary to maximize profit. State-of-the-art software platforms can help in achieving this goal through efficient data screening and optimization of multiple process parameters. In this case study, data reconciliation and multivariable optimization (MVO) were carried out on a DHT in Russia, using proprietary optimization software^a, resulting in increased earnings of $2.92 MM/yr.

DHT is a mature technology, and the process is well described in literature.¹ A simplified process flow diagram of the unit that has been studied is shown in FIG. 1. Equipment descriptions are provided in TABLE 1.

METHODOLOGY

Data reconciliation

Process measures can be corrupted by such factors as power supply fluctuations, network transmission and signal conversion noise, analog input filtering, changes in ambient conditions, improper installation of measuring devices, miscalibrations, wearing and corrosion of sensors, and plugging and deposition of solids. Data reconciliation aims at obtaining consistent estimates that will satisfy all the specified model equations, while staying as close as possible to the actual measurements. To understand the data reconciliation technique followed by the optimization software used in this study, it is important to define the term offset (Eq. 1):

        Offset = Model variable – Scanned (measured) value                                   (1)

Due to uncertainty in measurements, the measured data (X)—e.g., mass flowrates in and out of an operating unit—will not satisfy the principle of conservation of mass. The optimization software seeks to find an improved solution, x, that conserves mass and is acceptably close to the original measurement. In mathematical terms, the data reconciliation tries to find a solution for x (Eq. 2):

        Minimize (x conserves mass) ||X – x||²                                                      (2)

The function ||X – x||² is called objective function.² During data reconciliation runs, the optimization software keeps the scanned value constant and adjusts the model variables (FIG. 2) to minimize the overall objective function of the entire model—thus, bringing the model as close as possible to the actual operating conditions of the plant while maintaining mass, energy and component conservation.

Multivariable optimization

The aim of MVO in an industrial process is to optimize the plant parameters to meet a particular objective, which usually is maximizing the economic objective (i.e., profit). The MVO algorithm is based on equations developed by Zanin et al.^3,4 for real-time optimization of continuous processes. During MVO, the offsets calculated for each parameter in the data-reconciled model are kept constant, while both the scan value and the model value change. This ensures that, during MVO, the optimized model maintains the same level of closeness with the real plant operating conditions as achieved by data reconciliation (FIG. 3). The optimization objective—maximization of profit, in this case study—is defined in a multivariable controller (MVC), along with the parameters that need to be optimized.

The parameters that are selected for optimization can be divided into two categories:

1. Manipulated variables (MVs): These are usually a setpoint of controllers, and they behave as independent variables. The level (optimized) value of these variables are the new controller setpoints, which the operator needs to enter to achieve the objective (extra earnings/profit) calculated by the software.
2. Control variables (CVs): These are usually instrument measurements, slave controllers of a cascade control or product properties. Although the optimization of these parameters is important, they cannot be directly controlled and are dependent on the changes in MVs. These parameters are also known as dependent variables.

CASE STUDY

Data selection, screening and input

Hourly average data of instrument measurements were obtained from the refiner for the DHT unit for a period of 1 wk. The data was scanned to identify the steady-state moments of the unit, and one particular data set was selected for the data reconciliation exercise. Steady-state moments can be defined as the periods of operation where the plant is running smoothly at a more or less steady state (i.e., there is no ramp-up or ramp-down of the unit capacity, and no upset condition). The product and utility prices that were obtained from the refiner are provided in TABLE 2. Cooling water and lean amine prices were not provided by the refiner, and their contribution has not been considered.

The following constraints were defined in the optimization study:

• The reactor’s operating pressure, weighted average bed temperature (WABT) and treat-gas-to-oil ratio should not be changed.
• The kerosene product flowrate and properties (the ASTM D86 distillation endpoint and flashpoint) should not be changed.
• The diesel product should meet the following specifications:
  a.   95% recovery (v/v), ASTM D86 distillation (maximum): 365°C
  b.   Flash point (minimum): 55°C
  c.   Standard specific gravity: 0.82–0.845.

With these constraints, the economic maximization objective was targeted by the following routes to:

• Adjust the product draws (except kerosene) to maximize the production of high-value product diesel, while maintaining the required properties
• Optimize heat integration of the unit, so that the load on the fired heaters—and, consequently, the fuel gas consumption—can be minimized.

The optimization parameters chosen to meet these objectives are provided in TABLE 3.

Results and discussion

By data reconciliation, the model variables were brought as close to the scanned (measured) values as possible, while adhering to the principles of conservation. The data reconciliation resulted in a reduction of mass balance error from 3.5% to 0.06% (TABLE 4). The data-reconciled model was used for optimization.

TABLE 5 shows the initial and level (optimized) values of the optimization parameters. The level scan values of the MVs are the new setpoint for the controller under the optimized environment.

TABLE 6 shows the initial and optimized values of the products and utilities, along with the extra earnings that can be achieved by this optimization. As shown in TABLE 6, additional earnings of approximately $349/hr—or approximately $2.92 MM/yr, considering a plant uptime of 8,400 hr/yr—is possible with the given constraints. About 77% of the additional earnings are achieved by increased production of high-value product, while the rest is achieved by a reduction in utilities consumption.

By comparing the values of TABLES 5 and 6, it is clear that optimization has been achieved by the following routes:

• The fractionator (C-2) operating conditions—such as overhead temperature, pressure, top tray reflux flow and pumparound
  flow—were adjusted so that the draw of the high-priced diesel (a mix of heavy naphtha, light diesel and heavy diesel) is maximized at the expense of low-priced light naphtha.
• The heat-exchanger approach temperatures were tightened as much as possible with the existing arrangement, resulting in an overall reduction in fuel gas consumption.

Takeaway

Data reconciliation and MVO have been carried out on a DHT, using proprietary process optimization software. The results show that, under the given constraints, there is potential to increase the earnings from this unit up to $2.92 MM/yr by suitable adjustments of the operating parameters. The methodology discussed can be followed to optimize other units in an oil refinery and in different process plants in general. HP

NOTE

       ^a Refers to AVEVA’s SimSci ROMeo process optimization software

LITERATURE CITED

1. Meyers, R. A., Handbook of Petroleum Refining Processes, 4th Ed., McGraw-Hill, New York, New York, 2016.
2. Simulation Sciences Inc., Getting Started with ROMeo and ARPM, Lake Forest, California, 2008.
3. Zanin, A. C., M. Tvrzská de Gouvêa and D. Odloak, “Industrial implementation of a real-time optimization strategy for maximizing LPG in a FCC unit,” Computers & Chemical Engineering, July 2000.
4. Zanin, A. C., M. Tvrzská de Gouvêa and D. Odloak, “Integrating real-time optimization into the model predictive controller of the FCC system,” Control Engineering Practice, August 2002.

 

The Authors

Related Articles

From the Archive

Comments