Using artificial intelligence to reduce emissions

Model-based emission monitoring allows predict plant emissions in advance and take action before violations occur


ABB's Predictive Emission Monitoring System (PEMS) uses an empirical model to predict emission concentrations based on process data, and it has been successfully implemented as part of a comprehensive Environmental Management Systems in one of the largest gas processing plants in the world.

Role of modeling in emission management systems

A Predictive Emission Monitoring System PEMS – also known as an inferential analyzer – cannot measure emissions directly but uses an empirical model to predict emission concentrations based on process data, such as fuel flow, load, operating pressure and ambient air temperature. In fact PEMS can provide the only way of obtaining a continuous stream of (estimated) emission values in process units where well-known and reliable monitoring system is the Continuous Emission Monitoring Systems CEMS’ are not present and where either in-situ (ie, periodic) analysis or the campaign approach is implemented. In such cases the plant is allowed to lease a portable CEMS to gather sufficient emissions data to build and validate the models. Once the models have been certified, the CEMS is removed and replaced by the inferential type system. PEMS can also be used as a back-up if a CEMS is in place, and irrespective of which role it plays, it provides numerous benefits in different applications.
Acquiring proper, and reliable information about emission levels is crucial if adequate control actions to keep emissions inside law-enforced limits are to be deployed.

Main modelling approaches

Modeling is used to develop compact mathematical expressions that describe the behavior of a process or equipment. There are two main approaches: theoretical and empirical. A theoretical model is derived from scientific principles, such as the conservation of mass and energy, and the laws of thermodynamics, while an empirical model is mathematically derived from plant-specific process data. In general, modeling is able to provide an accurate real-time estimate of difficult-to-measure quantities; exploit otherwise hidden or neglected correlations; and provide a deeper insight into the process.

Estimated quantities are often referred to as inferential variables and the model is also called an inferential model. Advanced process control strategies usually employ inferential models. The relationship between the input data (ie, available measured variables) and output data (ie, the variable that needs to be estimated) is determined during the model building stage.

Dedicated software is used to import, pre-process and filter out historical datasets, which must include all the possible samples of the quantity that needs to be estimated. The resulting model has to be extensively tested and validated on the widest possible range of operative conditions. When this is completed, the model can be placed online where it is fed with real-time process data. This data is generally pre-processed to identify transient states and filter out possible outliers and bad qualities. The model output is also pre-processed to increase its reliability and accuracy.

Potential applications areas for AI technologies

Many applications have proven that software systems are just as accurate as the hardware-based CEMS. In addition, virtual analyzers offer other functionalities that can:

  • Identify the key variables that cause emissions
  • Automatically validate sensors
  • Reconstruct emission levels from historical data when the hardware device fails
  • Complement and enhance process optimization strategies

Actual regulation requirements insist that periodic tests need to be performed at the stack as well as continuous emission monitoring in order to prove compliance with the legal limits and track eventual violations. A conventional CEMS, however, cannot anticipate pollutant limit violation. A PEMS, on the other hand, could allow plant engineers to directly correlate the relationship between varying operational parameters, predict plant emissions in advance and take action to adjust emissions before violations occur.

The mood around the world regarding the methods used to monitor emissions is changing: Many European regulations now explicitly call for software-based redundancy emission monitoring systems while in the US, several states allow artificial intelligence (AI) technologies based on models like PEMS as an alternative monitoring technique.

First EPA validated system based on predictive technologies

ABB's innovative PEMS solution has been successfully implemented at one of the major gas providers in the Gulf region that produces network gas, natural gas liquids (NGL), condensate and sulphur. At this plant PEMS has a crucial role in emission monitoring as it works solely for gas turbines. To design the most appropriate model for PEMS, a temporary CEMS analyzer was used at each stack unit to acquire proper emissions data, while simultaneously process data were collected directly from the plant DCS through an OPC protocol. Data collection lasted about six weeks to cover the widest range of process conditions.

Model design and validation, data processing and site implementation activities were executed using ABB’s Inferential Modeling Platform. Data processing is a key step during the development of an empirical model. To begin with, both the input variable set and optimum pattern for the plant model were defined using sophisticated statistical and mathematical techniques. Identifying the optimal sampling rate for modeling purposes was critical because it has to satisfy two purposes: allow the identification of process dynamics and conditions; and provide an adequate number of suitable data sets to create good and accurate models.

The system was integrated with the DCS and the estimated emission values were configured to be written to the EMS via the serial protocol Modbus.

Results and achievements

Once installed, the system was subjected to an Environmental Protection Agency (EPA) assessment and certification process by an authorized third-party company from the United States. The process required 18 test runs lasting 30 minutes each at two different operating conditions (ie, nine at 95 percent of compressor load and the rest at 100 percent). After each test run, the emissions estimated by PEMS were compared to the values measured by CEMS, enabling the relative accuracy 1 of the PEMS system to be determined. As the performances for each emission were compliant with EPA regulation, the system was certified and then finally accepted by the customer
Property Ra 95% Load Ra 100% Load
Oxygen < 10% < 10%
NOx < 10% ≈15%
SO2 undetected (<1ppm) undetected (<1ppm)
CO < 10% < 15%
CO2 < 10% < 10%

The EPA assessment and certification process required 18 test runs lasting 30 minutes each at two different operating conditions

Advantages of easily adaptable “smart” system

The success of the first EPA validated system based on predictive technologies in the Gulf region has opened the way for further applications in that area. Many see the advantages of having an easily adaptable “smart” system:

  • Its performance is accepted by internationally recognized environmental agencies.
  • A predictive system can improve traditional CEMS availability.
  • PEMS can actually replace traditional analyzers in cases where CEMS are not available or usable.

In addition to these advantages, the simulation features provided by the ABB solution allows plants to investigate possible operation improvements in a non-invasive environment and determine best practices to run the process. They also enable the advance testing of optimization systems to satisfy local environmental regulations.

ABB flagship Inferential Modeling Platform software  has been successfully applied to provide highly reliable model-based monitoring and real-time estimation in power and process industries in five continents. ABB has installed model-based strategies for environmental purposes on Turbo-compressor stations, sulfur recovery units, FCC and other refinery units and even on a polymer plant.

Enabling technologies

Model Predictive Control demystified
  • お問い合わせ

    ご質問内容などをご入力の上、お送りください。 担当者より折り返しご連絡いたします。

Select region / language