Comparison of Forecast Accuracy between CSE ProdCast and Traditional Decline Curve Analysis (DCA) Model
Production forecasting is a critical activity for oil and gas companies, as it can significantly influence field development planning and the economic evaluation of oil and gas assets. Traditional methods employed for production forecasting, such as numerical simulations and decline curve analysis models (DCA) such as Duong, Fetkovich, MBT, Stretched Exponential, etc., are based on assumptions and require extensive domain knowledge. Moreover, for an unconventional reservoir, the estimation of future oil and gas production is complex due to the heterogenous nature of formations and dynamic operational events. Due to these complexities, combined with the rapid decline rate of wells, an unconventional well often does not follow the natural trend of the depletion regime.
To overcome these challenges, CSE ProdCast was built using an artificial neural network for predictive modeling using time-series data as a mechanism for predicting a more ‘natural response’ from the reservoir. This new approach leverages a data driven model that learns directly from data without making any prior assumptions. It provides an alternative quick, accurate, and cost-effective approach for forecasting production of existing and new wells. The solution can also be extended to build production forecasting models using data from multiple wells.
The biggest and most important advantage of using CSE ProdCast is the increased accuracy of the forecasting model as compared to the traditional decline curve analysis methods. The increased accuracy comes from the solution taking a data driven approach without making assumptions. The model is robust enough to incorporate any dynamic changes in operational parameters, such as temperature and pressure. The forecasting solution has the capability to forecast all three phases (oil, gas, and water) together unlike traditional methods which only forecasts a single phase. The solution is scalable to forecast the production of a well using aggregate data from multiple wells which renders the forecast more realistic and accurate. Finally, it is a quick and scalable solution.
CSE’s forecasting solution pulls time-series data directly from the OSIsoft PI System utilizing the Asset Framework (AF) SDK. This data is used to run advanced forecasting algorithms in real-time. The results from the algorithms are fed back to PI and can be displayed and compared to actuals using any of the off-the-shelf PI visualization tools. CSE’s solution can be installed on any Windows server where the AF SDK is installed that can be used to retrieve and write data to the PI. Examples include installing the solution directly on the PI Data Archive server, a server running other analysis services (e.g., the PI Analysis Service), or a dedicated server.