Abstract
Abstract
Energy companies' data bases hold the potential for tremendous advantage. Moreover, they are growing exponentially, fed by a mounting array of sources including ERP systems, sensory networks, machine instrumentation, credit card transactions, loyalty programs, and increasing multi-media and other unstructured data. Yet analyzing and extracting value from large datasets has always been a problem due to the sheer volume of data and consequent "signal to noise" problems.
To unlock some of this value, BP have launched a "predictive analytics" initiative, investigating pattern recognition techniques that find correlations and relationships in large datasets. This form of analysis is not new, but pilots in previous years have not resulted in significant implementation. However, recent advances in computing power coupled with more sophisticated applications, easy-to-use interfaces and easier access to data, have made predictive analytics a more valuable, practical tool for wide scale corporate adoption. And now, with more data available from sensor enabled applications, predictive analytics presents an opportunity to make all aspects of our field operations more efficient and effective.
A particular distinction of these techniques is that the models used are data-driven, and do not rely on a fundamental understanding of the first principles involved. This means that predictive analytics can be particularly useful where the underlying scientific principles are complex and difficult to model precisely. Data driven models can be created with reduced effort and often without the need for in-depth skills used to create models from first principles.
To take advantage of these techniques, the Digital & Communications Chief Technology Office (CTO) has been leading a program to introduce predictive analytics across all business segments and operations in BP. In so doing, we have been working closely with our segment specific technology teams, FIELD OF THE FUTURE, Refining Technology and our Exploration and Production Digital & Communications Technology teams. Having established connections with over 50 "predictive analytics" vendors, CTO is working with business units and functions and have completed a number of proof of concept (POC) trials utilizing a variety of innovative vendor solutions
This paper describes a number of these POCs and presents the value proposition for data driven predictive analytics.
Predictive Analytics
Why now? Using sophisticated algorithms to detect useful patterns in large amounts of data is not new. Data mining, as it is often called, has been around for a number of years and for some time now businesses have been employing a variety of mathematical techniques in attempt to gain useable insight into their data. However, previously, these analytical techniques were compute intensive and often required significant effort from statisticians and analytical experts. The cost, therefore, was not insignificant but perhaps more importantly, the whole process took time and the business would often have difficulty understanding the context of the results. The lengthy process often meant that, whilst the business may have gained insights into why something had happened, it was often unable to predict an event within a timeframe that allowed action resulting in business value.
So what has changed? One of the more obvious advances has been the increase in compute power available at significantly lower costs. But, in addition and maybe partly because of this, we have become aware that vendors are finding ways to incorporate many of the traditional mathematical algorithms for data analysis into software with relatively user-friendly interfaces. This type of analysis is not now merely the domain of the statistician but of the power business users, with operators and engineers now entering into the space. In some cases relatively new algorithms have been developed, patented and incorporated into software tools. It would seem that this all has led to a situation where businesses are now able to mine data to gain not just insights but the information from which to construct usable models. Models that are able to be used to predict future events within a timeframe that allows business decisions to be taken ahead of the event occurring.
Cited by
7 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献