logo


Oct-2016

Data analytics for hydrocarbon processing

Advanced data handling techniques are transforming operations in the refining industry at an increasing pace.

JOHN McCALL
The Data Lab

Viewed : 4779


Article Summary

Big data, data science, data analytics, the emergence of fast and flexible information systems, the internet of things and cloud computing have revolutionised the ways in which data can be accessed and used by industry.

Technology has provided an environment where data can easily be retrieved from a variety of sources and reassembled for analysis. This makes it possible to bring together and analyse data in ways that were previously impossible. The term Industry 4.0 is being used to describe how all of this is changing the way companies operate, often in fundamental and disruptive ways.

Companies involved in processing hydrocarbons already generate vast amounts of data through metering for allocation, process control and safety monitoring purposes. The industry is no stranger to analytic techniques and there is some automation in monitoring systems, with advanced techniques like fuzzy logic and neural networks used in alarm systems. The advent of modern data technologies therefore offers significant opportunities to realise additional value from data all along the production chain.

The term data analytics is very broad, covering a whole range of technologies. Some of these support the storage management and exchange of data. For example technologies such as Extensible Markup Language (XML) and JavaScript Object Notatation (JSON) allow the development of flexible data formats for particular domains which are very easy to automate. This greatly helps in bringing together previously isolated data sets. The formats are ‘self-describing’, which means that data, and standard analyses, can be made available through web services. This in turn makes it very straightforward to construct data pipelines that combine and move data through a series of analyses in a fast and automated way.

Once a pipeline is constructed, various analyses can be added. Many proprietary software platforms for general data analytics, for example TIBCO Spotfire, are available on the market and there are a growing number of specialised analytics platforms for particular industries. Another powerful development is open source community developed platforms. A leading example is the R community, which offers a vast array of ‘modules’ and ‘scripts’ all of which can plug together flexibly to create entire data pipelines from source to analytics to visualisation and reporting.
As an example, the R module ‘chemometrics’, developed by Kurt Varmuza and Peter Filzmoser of the Vienna University of Technology, offers a range of analytic methods widely used in chemometrics, including principal component analysis (PCA), principal components regression (PCR) and non-linear iterative partial least squares (NIPALS), as functions that can be called in any general R script. Combining modules such as chemometrics in scripts with other R modules for sourcing data and visualising results makes it possible to assemble powerful analytic tools with reasonably lightweight programming effort. R has become the leading platform for the data science community and, as of April 2016, offers over 8000 distinct and diverse analytic packages.

In this article we present two data analytics case studies, one from downstream processing and one from the upstream industry, which illustrate what can be achieved.

Data analytics in the upstream industry
Accord-ESL offers hydrocarbon allocation services to the upstream industry. Recently the company launched Charm (Compact Hydrocarbon Allocation Reference Model), a software simulation specifically designed for hydrocarbon allocation systems, which can be delivered as a web service.

The main purpose of simulation models, used within hydrocarbon allocation systems, is to provide information regarding how hydrocarbons are behaving in a process plant. The use of simulation models ranges from the prediction of 
physical properties and calculation of shrinkage factors, to full integration of the model (using cloned components) within the allocation process itself.

Traditionally, general purpose process simulation packages such as HYSYS and PRO/II have been used to calculate these process effects. Though at first sight there may be an appeal to using such models, there is a number of issues that mitigate against their use:

• Model complexity and stability: general purpose simulation packages are built to model a wide range of processes and are routinely used by process engineers for design purposes. Such models can be complex and include such items as control valves, pumps, compressors, heat exchangers, and so on, and consequently may frequently require intervention by a process engineer to ensure they solve.

• Software integration: neither allocation nor simulation software are developed specifically to be integrated with each another. The effort required to facilitate their smooth communication is often regarded as too difficult by the operators of allocation systems and hence such integrated systems are rare.

• Software updates: even if an interface between allocation and simulation software is successfully implemented by an operator, the vendors will routinely update their software, potentially rendering the interface between the two systems non-functional. Also, even minor modifications to simulation software (for instance changes to the solution algorithm) can result in small changes to allocation results, which is an undesirable feature of an allocation system, where repeatability is important.

As a result of these issues, operators often resort to alternative approaches, such as running the simulation off-line and only updating the factors from the model periodically. The reality is often that the models are not updated and the update frequency may be arbitrary in the first instance. The use of the model every day is far more desirable as it automatically captures changes in the process.

Another alternative is to curve-fit empirical correlations to simulation data in an attempt to mimic the simulation results. Such correlations are approximations of the simulation and can be poor if they are functions of multiple variables. In addition, they are only applicable over the range of conditions they are designed for, and if all potential production scenarios are not anticipated they will produce spurious results if applied outside their range of applicability.

For allocation purposes, there is a need for accurate process simulations that are seamlessly integrated within existing allocation software.

Accord has therefore developed Charm to bridge this gap. The software has been developed principally over a 12-month period in collaboration with Robert Gordon University. It provides the accuracy of the full blown commercial software, whilst concentrating on being readily integrated into existing allocation software packages.

How has this been achieved? Generally in an allocation system, the simulation is only used to determine how hydrocarbons entering the process are distributed between the various liquid and gas products exiting the process; stream enthalpies, equipment performances, and so on are not of interest. Charm only solves the equations necessary to determine how the hydrocarbons move through the process, using the same underlying equations to model the vapour liquid equilibria in the various separators as the commercial software packages, for instance the Peng Robinson equation of state.


Add your rating:

Current Rating: 4


Your rate: