Using Predictive Emission Monitoring Systems (PEMS) in Refining Processes

The on-going acquisition of emission data in process industries is regulated by legal requirements so that the release of pollutants to the atmosphere is monitored and controlled. The acquired data is a check on plant emissions to make sure that they do not exceed the threshold limits laid down by the law.

The availability of efficient and reliable emission data acquisition tools is critical for plant owners.

When there is extended failure on the part of a plant in providing emission values, legal authorities may impose a shut down on the plant. Moreover, environmental limitations affect the production of the plant.

The continuous emission monitoring systems (CEMS) used in most plants are hardware-based. The CEMS is made up of analyzers for sampling and identifying the composition of the flue gasses released, and an IT infrastructure for managing, recording and storing the emission values.

An effective alternative to CEMS is the predictive emission monitoring systems (PEMS) that is based on software and accepted by most environmental regulators for monitoring and recording plant emissions. PEMS is a new concept that estimates the concentration of emissions based on advanced mathematical models.

The empirical modeling technique, also known as inferential or data-driven modeling, is considered as the most effective technique for creating precise models for emission estimation.

This technique is based on the ability to derive relevant information from datasets to predict the behavior of pollutant concentrations, which depend on the physical variables that characterize the emmisions releasing process. By using Artificial Neural Networks (ANN), the performance and robustness of the model can be balanced in order to provide the reliability and accuracy comparable with that provided by hardware-based emission analyzers.

The details of a successful implementation of neural network technology in a major refining plant located in Southern Europe are provided in this article.

PEMS Rationale and Process Overview

Although PEMS has been accepted as a primary source for monitoring emissions by the US-EPA legislation, the European regulatory authority considers PEMS only as a backup for CEMS. Based on the legal regulations, a top European oil refinery opted to implement PEMS as a backup for the existing emission monitoring infrastructure that used CEMS.

The implementation aimed at increasing the service factor of the hardware analysis system above 97.5% and reducing the number of third party company interventions for monitoring emissions when the hardware analyzers were not in service.

The redundant values of various pollutant components acquired from the fluidic catalytic cracking (FCC) and sulfur recovery units (SRUs) by PEMS. These two units are critical and more complex than the units that are generally considered as the most suitable for PEMS implementations.

This refinery, in particular, used upgraded and modified versions of the FCC and SRU units in order to curb plant emissions and improve the refining capacity.

Sulfur Recover Units

The exhaust gases emitted by the three parallel desulfurization trains are directed to the SRU stack. Each of these streams implements different treatment methods and processing units, but the Claus process implemented downstream is virtually identical in all three trains. There are many bypass valves provided in these trains for selective diversion of the process gas (Figure 1).

Sulfur recovery units layout

Figure 1. Sulfur recovery units layout

The tail gas treatment units (TGTU) in the second and third trains are patented and different. The catalytic incineration stage follows the TGTU stage. The sulfur extraction in the first unit is less efficient because only a thermal incinerator is provided in it.

SRUs receive gasses from various types of refinery treatment and production units. The gases that were fed into the SRUs were of unknown composition and ratio and not fixed over time. A rich stream of gases like H2S, NH3 and CO2 of varying concentration is fed into the three streams of SRUs.

Fluidic Catalytic Cracking

Further treatment of the flue gas from the FCC regenerator was done by the patented absorption process to reduce the amount of SO2 emission. The new FCC unit (Figure 2) is provided with its own stack (FCC-02). The exhaust gas is diverted from the cracking unit either to the absorber or directly to the original stack (FCC-01).

FCC absorption units layout

Figure 2. FCC absorption units layout

The implementation of an effective predictive solution is complicated by the plant layout and processes. The first complication was due to the fluctuating composition of the feeds, which depends on the performance of the upstream units and the hydrocarbons that are processed by the refinery in the beginning. The operator has no control over these fluctuations. Another complication was due to the several operating scenarios of both the units:

  • In SRUs, the various sub-processes can be operated in numerous configurations, based on the variations in load and maintenance activities for generating different emission levels.
  • Compliance with environmental constraints is enabled by the SO2 absorption unit. In its active state, up to 50% of the FCC off gases are diverted to the absorber, and later to the FCC-02 stack. In the inactive state of the unit, all the gases are diverted to the FCC-01 stack.

The engineering phase was largely influenced by these operating conditions, and hence, an in-depth analysis of the process behavior and collaboration with the plant personnel was required for comprehensive assessment of the operations of the unit and instrumentation that was available.

PEMS Solution

By working closely with the refinery engineers, the PEMS team was able to clearly define the standard operating conditions that were needed for developing the system. The PEMS application for the SRUs was customized to deliver best performance in the most general scenario, in which the highest sulfur removal efficiency was achieved. In both scenarios, the TGTU2 and TGTU3 operated with a tail gas that was diverted from the first unit to TGTU2.

In the case of cracking units, the software analyzers were designed so an accurate measurement was provided for both stacks, based on the valve open-position value that helps possible shutdown of the SO2 absorber. Creation of a dataset that represented a set of variables that covered the standard operating conditions and described the process dynamics was essential for building an effective model.

The first step was the data-acquisition phase where a baseline was collected of time-stamped and synchronized emissions and process data which was used for creating the model. This step involved the extraction and analysis of six month old data that was stored in the plant emission data acquisition system. The final subset of variables that were used for developing the model was derived by subjecting the initial dataset to several operations:

  • Removal of outliers and data of ‘bad quality’
  • Identifying the proper sampling time so that the model overtraining and the loss of important data on process variability are balanced
  • Statistical analysis conducted by advanced mathematical techniques like principal component analysis help reveal the hidden correlation between the process parameters and emission values

By following the steps listed above, PEMS engineers were able to narrow down on the operating parameters that are suitable to be used as input variables. Since the SRU models involved a large number of units, they required a set of 10-12 input parameters on an average for achieving proper accuracy. The cracking unit required only seven or eight input variables. To identify the right model for reproducing accurate emission values, a number of model structures like linear regressions, neural networks, etc, were generated and their performances were compared.

Based on these comparisons, the engineers chose neural networks as the appropriate model architecture because of its robustness and effectiveness in emission monitoring.

Feed forward and neural network schematic

Figure 3. Feed forward and neural network schematic

Once the off-line validation was completed, software analyzers were installed on a dedicated server that was located on site. Later, an OPC-connection was setup so that the PEMS software engine received the real-time process values from the control system.

Real-time emission estimations were produced based on the parameters processed by the model. Finally, the PEMS system was integrated with the existing data acquisition system (DAS) for making the emission data accessible to plant personnel (Figure 4).

System architecture schematic

Figure 4. System architecture schematic

In scenarios where the traditional instrumentation is not provided data, a strategy was devised to extend the PEMS values to the emission ‘bubble’ limit of the refinery.


The PEMS estimations were validated by comparing the values produced by the system with the values provided by the existing hardware instrumentation for attaining final acceptance by the refinery. The comparison results showed that the predictions provided by the software analyzers were in excellent correlation with the results from the analytical devices.

The predicted SRU flow values plotted against the real-time data obtained from the flowmeter that is mounted on the stack are shown in Figure 5.

PEMS vs CEMS for flue gas flow at SRU stack

Figure 5. PEMS vs CEMS for flue gas flow at SRU stack

According to the figure, the PEMS values are in line and are within the ±5% bandwidth from the physical measurement over the 20-days period reported.

The significance of the PEMS implementation lies in increasing the total availability of the emission monitoring infrastructure on site. In case of CEMS, during the normal maintenance periods, inferential models provided redundant data that helped cover the blank periods.

A daily chart depicting the predicted and measured NO emission values at FCC stack are shown in Figure 6.

PEMS extends emission monitoring availability

Figure 6. PEMS extends emission monitoring availability

As a result of daily automatic recalibration and maintenance activities, there were no emission measurement values from the hardware analyzers during two distinct intervals. This scenario was overcome by PEMS, which provided an alternative measurement and also an overall service factor well above 99% for the emission monitoring infrastructure.


Based on the results, software analyzers are able to provide a highly accurate solution that is a dependable backup to CEMS systems in a highly challenging refinery process. For such applications, any mismatch in the output from the PEMS model and the analytical measurements could trigger early warning of malfunction or drift of the hardware, aiding maintenance action.

Compared to traditional CEMS systems, predictive systems provide well-trained inferential model that enables plant operators to execute off-line simulations of the emission behavior for different operating conditions.

With the help of ‘what-if’ analysis, plant engineers are able to examine the response of emissions to variations in the input variables and  how each operating parameter affects the final emission values. The contribution of the PEMS solution is much more than just a backup for CEMS. It can be implemented as a primary monitoring technology for numerous applications, proving its capability of providing accuracy and performance that can be compared with conventional analyzers.

On the economic front, the PEMS is much more advantageous than conventional analyzers, right from the initial capital to the operating costs. Some of the other advantages include:

  • Does not require any preventive of periodic maintenance
  • Near zero power consumption
  • Does not require any spare parts or consumables, reducing the warehouse necessities

Based on the above benefits and others, implementation of the PEMS serves to reduce the overall life cycle cost by 50%, when compared with conventional analyzers. Even though the predictive systems are able to deliver excellent results, they cannot replace CEMS. The choice of the system depends on the equipment, process layout and operative conditions. An integration of both hardware and software emission monitoring strategies is required for an effective solution that covers the entire range of possible applications.

For instance, PEMS may be a better process for boilers, furnaces or gas turbines, while CEMS would suit civil incinerators or in units where solid fuels are burnt. It is always better to depend on a supplier who exhibits solid capabilities and background in both these technologies in order to receive proper guidance in identifying the appropriate technology for the particular application.

Gregorio Ciarlo

Figure 7. Gregorio Ciarlo

Federico Callero

Figure 8. Federico Callero

This information has been sourced, reviewed and adapted from materials provided by ABB Measurement & Analytics.

For more information on this source, please visit ABB Measurement & Analytics.


Please use one of the following formats to cite this article in your essay, paper or report:

  • APA

    ABB Measurement & Analytics. (2023, April 07). Using Predictive Emission Monitoring Systems (PEMS) in Refining Processes. AZoM. Retrieved on May 19, 2024 from

  • MLA

    ABB Measurement & Analytics. "Using Predictive Emission Monitoring Systems (PEMS) in Refining Processes". AZoM. 19 May 2024. <>.

  • Chicago

    ABB Measurement & Analytics. "Using Predictive Emission Monitoring Systems (PEMS) in Refining Processes". AZoM. (accessed May 19, 2024).

  • Harvard

    ABB Measurement & Analytics. 2023. Using Predictive Emission Monitoring Systems (PEMS) in Refining Processes. AZoM, viewed 19 May 2024,

Ask A Question

Do you have a question you'd like to ask regarding this article?

Leave your feedback
Your comment type

While we only use edited and approved content for Azthena answers, it may on occasions provide incorrect responses. Please confirm any data provided with the related suppliers or authors. We do not provide medical advice, if you search for medical information you must always consult a medical professional before acting on any information provided.

Your questions, but not your email details will be shared with OpenAI and retained for 30 days in accordance with their privacy principles.

Please do not ask questions that use sensitive or confidential information.

Read the full Terms & Conditions.