Knowledge layer is next ‘big innovation’

Upstream oil and gas operators are increasingly challenged to maintain the appropriate level of technical expertise to operate their assets in an optimized way. Seasoned engineers retire or change positions or employer. As routine work has gotten mostly replaced by computers, the human input into processes is getting more and more complex. Hence, losing people means losing knowledge.

The industry’s proprietary technical and business software applications enable neither knowledge capture nor ownership and intellectual property over the operator’s workflow solutions. Organizations still struggle to come up with a software landscape that enables the effective processing of cross workflows that go beyond the borders of a department or a technical discipline. This landscape of isolated software silos impedes scalable system integration; thus information technology departments struggle to implement service-oriented architecture.

Nevertheless, well and reservoir surveillance software solutions able to detect unfavorable operational settings and identify underperforming wells are available on the market. But to improve decision-making in asset operations, expert or advisory systems representing industry and operator’s best practices are needed.

FIGURE 1. The adaptive advisory system captures knowledge from experts and experiences to provide a framework for an intensive knowledge exchange. (Images courtesy of Myrconn Solutions)

Improved decision-making in asset operations

Operations in oil and gas fields are typically driven by the objective to maximize ultimate recovery or the recovery to a certain date. Production losses mean that a significant amount of money is left on the table. Hence, deferred production or underperformance must be detected as soon as possible after occurrence (or ideally even before either occurs). An activity needs to be initiated immediately to reduce the amount of lost production or to avoid any losses at all.

With regard to producing hydrocarbons in an optimized fashion, there are four main steps in related decision-making processes:

Free to focus: Data screening (e.g. pattern recognition enabled by special visualization and data mining) is used to identify symptoms that indicate that asset performance is not as good as expected. Patterns among wells are detected to identify similar behavior and reduce the complexity of the screening problem from several hundred sensors to a few categories of similar measurement types.

Truly understand the challenge: Petrotechnical analysis methods (e.g. sensitivity analysis on numerical or analytical models) are applied to identify the root cause of why the performance is below expectation. The objective is to identify the constraint, such as whether the liquid production is limited by reservoir deliverability, well production potential, or facility processing limits. Due to the ambiguity of some of the symptoms as identified in the first step, the outcome of this step will be probabilistic, indicating most likely causes but also possible alternative causes.

Improve decisions: Based on a definition of utility (e.g. maximize production, minimize losses, increase net present value, reduce lost time, etc.) decisions are suggested to solve the problems as identified in the first two steps. Previous experience from the same reservoir, similar situations in other reservoirs, or information from case studies are the sources needed to select the most promising action with regard to the utility given the constraint as

identified above. A definite selection will not be possible in this step. Therefore, the suggestion will be of probabilistic nature.

Increase knowledge: The impact of the actions resulting from the decisions in the third step is analyzed and verified. Did the action yield the expected results, or is the performance different from what is expected? The discrepancy between the expectation and the actual observation is the learning opportunity, which needs to be recorded, explained, and finally generalized to clearly identify whether or not this particular piece of information is applicable to a single situation, the whole field, or the whole company. The gained knowledge is stored and updated in the knowledge layer for future application.

The adaptive advisory system

There is a business need to detect events such as severe underperformance in the asset and to react properly and in as timely a fashion as possible to keep production up to target. When combined with the incomplete and uncertain information available from the sensors in the facilities and wells, this can lead to the fact that operations are very often cases of firefighting and rushing from one event to the next.

FIGURE 2. A Bayesian analysis factors in many datapoints to determine the most likely cause for declining pump performance.

Actions are usually taken reactively after a certain event has been observed. Moreover, the actions that are considered after a certain event has been detected often are not based on the full amount of technical expertise available to an oil and gas producing organization at a current time but rather on typical approaches and rules of thumb that have been around in an organization for ages.

The standardization of processes, common performance metrics, reporting, and documentation is hardly ever in place. Hence, it becomes difficult for an organization to monitor and support its producing assets and almost impossible for it to efficiently share information from one organizational unit to the next.

The adaptive advisory system is designed for an intensive interaction. It captures knowledge from experts and experiences and at the same time provides a framework for an intensive knowledge exchange of engineers, experts, and managers (Figure 1). It facilitates the asset team in decision-making processes and helps to file knowledge in a way that makes it accessible in future times as well as to other units.

Investigating the root cause of events involves further analysis of the available evidence and possibly the request of additional evidence. The outcome of the root cause investigation is a definition of the problem that has occurred in the asset. Therefore, it is a very analytical step where the user is requested to interact with the system and the data to confirm and investigate the root causes that led to the event.

A Bayesian network-based expert system

To set up a problem classification advisory system, the expert system can either be trained using historical event data or, where no sufficient historical data are available, the network structures and according probabilities can be determined by experts.

An example of root cause identification using a Bayesian network is displayed in Figure 2. The most likely root cause for declining pump performance is determined based on the Bayesian network taking into account the observations and real-time measurements. The observations can be independent (e.g. pump age and reservoir pressure) but also can be linked by a causal relationship (e.g. wellhead pressure and production rates). The example shows that given the information about the pump, reservoir, and fluid and measuring dropping liquid rates and wellhead pressures, the most likely cause is gas ingestion. However, mechanical problems and possible excessive pump wear (e.g. due to sand in the pump) should not be entirely excluded.

The added value

The knowledge layer approach helps operators overcome the dilemma of permanently losing knowledge by capturing it and building adaptive advisory systems. True added value is gained when a knowledge layer is incorporated in collaboration portals and business process management systems. Decisions will be continuously improved, and the advisory systems become increasingly more solid with a growing knowledge base.

References available.