A framework for timely and accessible long-term forecasting of shale gas production based on time series pattern matching

Abstract

Shale gas production forecasting is an important research topic in the gas industry. A common shale gas block includes dozens or even thousands of wells and therefore has a great number of historical production series. However, most existing methods apply single-well modelling. This cannot exploit data from other wells and requires a long production history from the target well, so the forecasting timeliness is compromised. Moreover, the parameters required by many of the existing methods are difficult to collect in practice, so the forecasting accessibility is compromised. Therefore, this study presents a shale gas production forecasting framework with improved timeliness and accessibility. To ensure timeliness, the proposed approach utilises historical data from existing wells and only requires a short production history from the target well. To ensure accessibility, the proposed approach only requires past daily production time and gas yield. The performance of the proposed method is demonstrated through a comparison with baseline methods. The results regarding cumulative gas production forecasting indicate that the proposed method has an average overall mean absolute percentage error (OMAPE) of 0.210, outperforming an artificial neural network with an average OMAPE of 0.241 and ARIMA with an average OMAPE of more than 2.