Abstract
Wells with a short producing life are particularly difficult to forecast reliably. Given the scatter in production data, acceptable decline curves can be fit with “hyperbolic b” values ranging from 0 to 4. These obviously result in widely different forecasts. The objective of this research was to develop a methodology to improve the forecasting reliability of wells with a short production life (less than 1 year), using “clustering” and knowledge gained from wells in the surrounding area
The production rate of each well was non-dimensionalized (from 0 to 1) with respect to its peak rate, and time was normalized to start at the peak rate. Each play was divided into a number of approximately equal sized areas. For each area, K-means clustering on time series was applied, and wells with similar decline trends were combined into clusters. The number of clusters in each area was dynamic, and depends on the number of wells in the selected area and the variety of decline trends. Type wells were generated for each cluster by averaging monthly production data of wells with long-production periods.
To test the methodology, 11000 wells with a long production life were selected. Two analyses were performed: a) individual wells were analyzed and forecast without the benefit of the auto-clustering methodology; b) the auto-clustering methodology was applied to create the forecast. The results of a) and b) were compared to the actual production of the long-producing wells. It was found that the results were improved significantly when the clustering methodology was used. An unexpected result of this study was that subdividing the Barnett Shale play into 9 areas and each area into several clusters, gave substantially the same result for a representative b-value as treating the total Barnett play as a single area. Similar results were obtained for the Eagleford and Bakken.
1. Introduction
The hyperbolic equation (Arps, 1945) is widely used to curve-fit historical production data of oil and gas wells. The strengths and limitations of this equation are well known in the industry. Notwithstanding its limitations, it is the most accepted method of forecasting future production of a well. Except for one topic, which is the subject of this research paper, the strengths and limitations of the hyperbolic equation for forecasting will not be discussed here.
