f From History to Prediction -Techniques for Conditioning Reservoir Models to Dynamic Data

Reservoir simulation models should always be built for specific business goals. It is an accepted rule that models used for production forecasts should reproduce the production on history. Although, most history matching processes are often the result of a complex team effort, objectives for using simulation models and the required level of detail are quite diverse. Applications range from prospect evaluation with limited available calibration data to designing detailed production planning scenarios for mature fields with highly constraining well production data. In either case, applied techniques, workflow requirements and the level of complexity will naturally differ. In recent years assisted history matching techniques and optimization workflows have been established and included in best-practice guidelines in an increasing number of companies in the oil and gas industry. The application of assisted history matching techniques is often motivated by the need to handle increasingly complex problem statements as well as the desire to improve workflow efficiency and transparency. Initially, the focus was given to finding single best models. Modelling paradigms, however, are changing. More recently, the industry has given a stronger interest to understanding a distribution of alternative scenarios which more realistically captures the uncertainty-envelope. This step is non-trivial, since there is no natural extension from the paradigm of single best history-matched models with deterministic forecasting capabilities to the paradigm of establishing a distribution of alternative production forecasts. This defines a major challenge to the reservoir engineering workflow and the question of handling multiple models with alternative outcomes. This talk reviews selected techniques used in history matching workflows. It discusses practical considerations for finding a compromise between “accurate” history-matched models with deterministic forecasting capabilities and the newer paradigm of a sufficient coverage of the uncertainty space for establishing uncertainty distributions.<br>