Optimize reservoir management in real time  
Real-time optimization  technologies and hybrid self-learning models are poised to meet reservoir  management challenges.

By David Wood, David Wood &  Associates; Saeid Mokhatab, University of Wyoming

Good petroleum reservoir management focuses on  maximizing profitability, production and ultimate reserves recovery by  integrating technical, commercial and risk management in a dynamic and uncertain  environment. Constraints (technological, physical, financial, geopolitical,  environmental, safety, corporate and human) limit the decision options  available.  

 Figure 1. Traditional framework of oil  and gas data sources, technical, economic and risk interpretation models and  decision-making tools. (Images courtesy of the University of Wyoming)    

Deploying new and better technology reduces  constraints and provides more and better quality information about the reservoir  that ultimately should improve reservoir performance.

Intelligent or “smart” technologies deployed in  well bores aim to provide better remote well performance monitoring and identify  early when reservoir interventions are required. Optimized well performance  means less lost production through downtime or inefficient operating conditions  and more profitability from production operations. The economic viability of  such technologies requires careful cost vs. benefit and risk analysis (Figure         1). In remote field developments they can  provide extremely cost-effective solutions, particularly where the information  they provide can be accessed and interpreted remotely from the well site.  However, such tools have yet to reach their full potential as the wealth of data  they provide is not processed and acted upon quickly enough. This is set to  change with the development of dynamic real-time optimization reservoir models  that should complement the more cumbersome traditional full-field multiphase  reservoir simulation.

To achieve maximum impact, smart technologies  should be deployed and the information they provide integrated with standard  reservoir and production management tools, databases and models to contribute  information to guide (and be guided by) real time reservoir optimization models.  

 Optimized reservoir management  

 Optimization of oil and gas assets,  either for profitability, cost, production or reserves recovered usually  combines mathematical models, field data and experience to influence investment  or methodology decisions. For example, a reservoir simulation model fed with  up-to-date reservoir information from well testing and downhole tools provides  multiple future production and reserves recovery scenarios based on existing and  carefully placed future wells (production and injection). Decision makers expect  reservoir managers to select the best options from the multiple simulation  scenarios. Optimization software can help reservoir managers to do this and  monitor performance of selected options as new production and reservoir data is  collected.

 If a reservoir simulation model is not  frequently updated with new data and new history matches conducted, the  “optimized” solutions rapidly become irrelevant and may then be put aside to  return to traditional tried and tested decline-curve and water-cut analysis,  which provide the reservoir managers with an understanding of what is actually  happening in active wells but offers little in terms of optimization solutions.  

 Reservoir models and mathematical  optimization routines are not one-off exercises but need  

Figure 2. Production and reservoir  dynamic real-time optimization model methodologies.

to be repeated frequently and acted upon when  new unplanned conditions prevail and production data becomes available. One  problem is that updating the reservoir simulation history-match can be a  laborious task unless data is provided systematically and in an appropriate  format. Unwieldy and slow integration of data into many reservoir simulation  models to ensure the adjustment of model metrics to match observed production,  water-cut and reservoir-pressure history commonly make the reservoir simulation  unsuitable for accurate short-term predictions and decisions.

 Reservoir models that can make  reliable short-term predictions are essential for production-related optimal  decision making. The increased availability of real-time data in the field  derived from intelligent completions can provide the necessary information to  feed short-term, data-driven reservoir optimization models. Indeed, permanently  instrumented wells that can be remotely actuated and interrogated facilitate  reservoir optimization performed in real time based on easy-to-manipulate models  updated by regular feedback from active wells.

Real-time optimization

Real-time optimization (RTO) is a  method frequently used in the downstream industry for complete or partial  automation of the process of finding good (optimal) control settings. By  continuously collecting data from a process plant, the data are analyzed and  optimal plant control settings are found. These settings are then either  implemented directly (closed-loop) in the plant or they get presented to an  operator (open-loop) for interpretation and a decision to implement or not. The  main aim of RTO is to improve utilization of the capacity of a production plant  to get higher throughput and improve efficiency and profitability. The model is  then continuously updated with new plant measurements and the best fit of the  actual input-output behavior of the plant is repeatedly recalculated and an  optimization control directive is issued.

This RTO approach can be adapted to optimize  performance from an oil or gas reservoir, well production and field process  facility complex. A general RTO system used in downstream plants consists of a  five-step procedure:

＆#8226;    Data validation;

＆#8226;    Model updating;

＆#8226;    Model-based optimization;  

＆#8226;    Optimizer command evaluation; and  

＆#8226;    Decision to adopt or reject  optimized solution.

 Self-learning models involve processes  by which a system uses its own past operating data to progressively further  develop and refine evolving algorithms as each new batch of new data becomes  available.

Integrating field data for continuous learning  of key reservoir features based on simplified hybrid models and multilevel  optimization is more suitable for real-time operations than full field-wide  optimization. Multilevel decision-making has also been developed and extensively  tested in the refining and petrochemical industries and can be adapted to  reservoir optimization.

 Self-learning models

Hybrid, self-learning reservoir models  are being developed and exploited when data are scant, as is often the case in  practice early on in reservoir development. These can balance the accuracy of  data fitting with the model’s predictive ability by appropriate selection of  model algorithms. Hybrid models may employ a first-principles structure along  with empirical constitutive equations (e.g., Darcy’s law, ideal gas law,  pressure-drop relationships) and rely on incoming data to identify and regularly  update values of many of the algorithms’ parameters. Because of this, hybrid  models are often easier to develop and manipulate than raw first-principle  simulation models, while maintaining model fidelity outside the range of the  data used for model parameter identification.

The desired model structure is a  self-learning adaptive scheme that optimizes multiphase fluid migration in  compartmentalized reservoirs, while integrating downhole completions, wellhead  restrictions and business constraints. It should continuously optimize reservoir  performance while satisfying surface and sub-surface constraints.

Various proposed RTO methodologies involve  dynamic models for short-term forward planning. Highly variable and  hard-to-process feed data arriving from the wells equipped with intelligent  completions make it hard to adapt steady state RTO solutions from the downstream  process industry. It is the dynamic nature of the model types described and the  multilevel optimization that allows them to process, learn from and act upon  uncertain and varying feed (Figure 2).

By continuous processing of data being  delivered from remote sites, RTO systems can identify and respond quickly when  plant and well go offline or move outside normal operating conditions. This  could have huge benefits in potentially preventing hazardous outcomes and  improving safety and environmental management. These benefits and the reservoir  management benefits all progressively reduce uncertainty, which should  ultimately lead to reduced operating costs.