A Novel Workflow for Oil Production Forecasting using Ensemble-Based Decline Curve Analysis

Abstract

In the absence of well-developed calibrated geologic and simulation models, empirical approaches such as decline curve analysis (DCA) are normally used for production forecasting and reserve estimation. DCA is computationally more efficient compared to simulation models when the active well base exceeds hundreds of wells. However, the underlying assumption for conventional DCA is no change in well operation settings. Moreover, the common approach for production forecasting consists of manual outlier detection and removal, interpretation of missing measurements and data fitting using different models for each well. Therefore, the process of conventional DCA is subjective due to the lack of a standard workflow for preprocessing and data cleansing. The common practice for doing DCA has three main steps: 1. Finding the most representative period in the history of well, 2. Detecting the initial rate (start point) of forecast, 3. Selecting the type of decline and fitting the appropriate model to data points. The solutions to these problems could vary from engineer to engineer and it can be time consuming to analyze all wells manually. To address these issues, we developed a novel workflow based on stochastic methods for detecting various well interventions including change in artificial lift, pump changes and acid treatment, and for forecasting oil production rate more accurately in the presence of uncertainty. The novelty of the proposed ensemble-based approach is forecasting conditioned on various well interventions. Furthermore, the proposed unsupervised stochastic anomaly detection method will detect various well works (or events) in the case of missing records of time and type of events. In this paper, we designed two experiments to test the proposed workflow for oil production rate forecasting and evaluation of acid treatments.