Advanced Analytics and Machine Learning Leads to Significant Return on Investment

Once OilCo developed its smart ОТ infrastructure across their value chain, with the EIDI as the centerpiece, OilCo's focus turned to machine learning and big data analytics. They became one of the first large refiners to adopt Microsoft's Azure Machine Learning in a production environment. Microsoft Azure works in conjunction with their EIDI (PI System) operational data, which is uploaded to the cloud using OSIsoft's standard software integrator for offline business analytics. This integrator made it much quicker and easier for OilCo to extract, cleanse, and contextualize their EIDI data so that it could be quickly ingested as published data sets by Azure's Power BI and Machine Learning cloud-based analysis tools. A data flow diagram for a typical integrated oil and gas company is shown in Figure 8.2, with data sources on the bottom and business use cases on the top of the diagram.

OilCo developed Azure Machine Learning to predict the impact of sulfur levels in various feedstocks in their desulfurization units. OilCo had been using offline models to analyze the sulfur content. Not only did using offline models increase time, it also increased the potential for error. OilCo estimated it was losing more than US$500,000 annually because of its inability to adjust unit parameters that optimize sulfur content in products. OilCo eliminated those losses, thanks to better forecasting, and continues to hone its use of machine learning. The company has rolled out this technology across their enterprise for other machine learning use cases. As with its other

FIGURE 8.2

Data sources and integrated analytics with an EIDI as the pivotal component.

improvements, OilCo was able to leverage its previous technology investments and reuse EIDI data, so that this new application was layered on top of what had already been implemented.

Following these successes, OilCo turned to improving the performance of its delayed coking units (DCUs). By using opportunity crudes, the company estimated that it could realize a gain of US$6 million for each 1% gain in DCU yield. Gains in DCU yields with variable feeds from opportunity crudes, however, also increased the risk of steam explosions during the hydrocutting step.

Azure analytics, combined with continuous EIDI data feeds enabled OilCo to find the sweet spot. DCU yields were increased by 2%, yielding an estimated annual gain of more than US$500,000 per year for each unit. Concurrently, steam explosions were reduced by 75%. Machine learning enabled the achievement of two seemingly contradictory goals at the same time. OilCo positioned machine learning for its four DCU units across the enterprise to take full advantage of opportunity crudes.

 
Source
< Prev   CONTENTS   Source   Next >