EXPLAINABLE ARTIFICIAL INTELLIGENCE (XAI) - ONE OF THE MAIN CHARACTERISTICS OF PETROLEUM DATA ANALYTICS (PDA); Section - 3

Part 3 of XAI: Type Curves

As was mentioned earlier in this article, Type Curves that are generated using mathematical equations are very “well-behaved” (continuous, non-linear, certain shape that changes in a similar fashion from curve to curve). Figure 16 demonstrates few more examples of Type Curves that have been generated in reservoir engineering. The question is, “what is the main characteristic of a model that is capable of generating series of well-behave Type Curves?” The immediate, simple answer to this question would be: “the model that is capable of generating a series of well-behave Type Curves is a physics-based model developed by one or more mathematical equations. The well-behave Type Curves that clearly explain the behavior of the physics-based model are generated through the solutions of the mathematical equations.”

The next question then would be: “What if it can be shown (and proven) that the model that has generated the well-behave Type Curves has not been developed using any mathematical equations? What if this model is purely data (field measurement)-driven?” Then the answer that would make sense would be: “(a) generation of these well-behave Type Curves proves that the data (field measurement) that was used to build the purely data-driven model reasonably represents the physical phenomena that was being modeled, (b) the data-driven predictive model can be clearly explained, and (c) the technique that was used to develop such model must be a scientifically solid technology.”

The main characteristics of the well-behave Type Curves is their explain-ability of the model that has generated them. Figure 17 and Figure 18 are good examples of “explaining” the model that has developed them. Figure 17 includes three series of graphs/Type Curves. These Type Curves have “Stimulated Lateral Length (ft.)” as their x-axis, and “6 Months Cumulative Gas Production (Mscf)” as their y-axis. Each of the curves in a graph represent the number of Hydraulic Fracturing Stages that are used during the completion of the well. There are six curves in each graph. Each of the graphs explains the behavior of a specific Marcellus Shale well in Southwestern Pennsylvania. The six curves in each graph show how each well’s productivity changes as a function of “Stimulated Lateral Length (ft.)” (from 3,000 to 10,000 ft.) and the Number of Stages (from 15 to 40 stages). The three wells that are shown in Figure 17 belong to three different pads in three different locations in this specific Marcellus shale asset.

Explainable AI Model for Unconventional Reservoir – Shale Analytics

Shale Analytics is the engineering application of Artificial Intelligence and Machine Learning in unconventional reservoirs. During Shale Predictive Analytics, that data-driven models that are developed for completion and production optimization are purely based on field measurements. In Shale Analytics use and incorporation of any types of mathematical equations are avoided due to the lack of realistic understanding of the physics of fluid flow in shale plays and the shape and characteristics of the fractures that are created as a function of implementation of hydraulic fracturing in natural fracture plays.

Once Shale Predictive Analytics is completed, through generation of Type Curves, Explainable AI (XAI) is used in order to explain the physics of the completion and production in shale wells. Such explanations that are based on actual field measurements and completely avoids any kind of assumptions, simplifications, and biases cannot be done through traditional approaches that have been used in our industry during the past decade. Traditional modeling approach of hydrocarbon production from shale wells using Rate Transient Analysis (RTA) and Numerical Reservoir Simulation (NRS) include minimum amounts of actual field measurements and are fully controlled by soft data. Soft data that fully controls RTA and NRS include Fracture Half-Length, Fracture Height, Fracture Width, Fracture Conductivity, and even Stimulated Reservoir Volume (SRV).

Using these soft data that can be generated by us and can have any values that we like them to have, instead of actual field measurements that are the main foundation of Shale Analytics, allow us to make any conclusions that we like to come up with even if they are 100% different from one another. In other words, such techniques can generate any kind of solutions that we like to come up with and have absolutely nothing to do with reality. It has to do with our objectives and has nothing to do with the realities associate with hydrocarbon production from shale wells.

Figure 17 and Figure 18 show how Explainable AI (XAI) can provide information for every single well in a Marcellus shale asset. Similar Type Curves can be generated for each pad, any specific part of the shale asset, or for the entire shale asset. In Figure 17 the Explainable AI (XAI) shows that the 6 Month Cumulative Gas Production of the Well #015 can increases from 850,000 MSCF to 1.3 million MSCF as the “Stimulated Lateral Length (ft.)” of this well increases from 3,500 ft. to 10,000 ft. This increase of Gas Production as a function of Stimulated Lateral Length is non-linear and can change the way it is shown in this figure when the completion design of hydraulic fracturing changed from 15 Stages to 40 Stages as long as all other variables that have been used to build this model will remain the same for this specific well. If the completion design includes 15 Stages, then the Gas Production can increase from 850,000 MSCF to 1.05 million MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft.

If the completion design includes twice as many Stages (30 Stages), then the Gas Production can increase from 900,000 MSCF to 1.25 million MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft.

For Well #112 (the middle graph in Figure 17) the Type Curves explains that 6 Month Cumulative Gas Production can increases from 100,000 MSCF to 900,000 MSCF as the “Stimulated Lateral Length (ft.)” increases from 3,500 ft. to 10,000 ft. This increase of Gas Production as a function of Stimulated Lateral Length is non-linear and can change the way it is shown in this figure when the completion design of hydraulic fracturing changed from 15 Stages to 40 Stages as long as all other variables that have been used to build this model will remain the same for this specific well. If the completion design includes 15 Stages, then the Gas Production can increase from 100,000 MSCF to 700,000 MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft. If the completion design includes twice as many Stages (30 Stages), then the Gas Production can increase from 100,000 MSCF to 820,000 MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft.

For Well #002 (the bottom graph in Figure 17) the Type Curves explains that 6 Month Cumulative Gas Production can increases from 750,000 MSCF to 1.25 million MSCF as the “Stimulated Lateral Length (ft.)” of this well increases from 3,500 ft. to 10,000 ft. This increase of Gas Production as a function of Stimulated Lateral Length is non-linear and can change the way it is shown in this figure when the completion design of hydraulic fracturing changed from 15 Stages to 40 Stages as long as all other variables that have been used to build this model will remain the same for this specific well. If the completion design includes 15 Stages, then the Gas Production can increase from 750,000 MSCF to 1.16 million MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft. If the completion design includes twice as many Stages (30 Stages), then the Gas Production can increase from 1.0 million MSCF to 1.21 million MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft.

Type Curves in Figure 18 explain that in this Marcellus shale asset, 6 Month Cumulative Gas Production of the Well #325 can increases from 0.1 million MSCF to 1.2 million MSCF as the “Stimulated Lateral Length (ft.)” of this well increases from 3,500 ft. to 10,000 ft. This increase of Gas Production as a function of Stimulated Lateral Length is non-linear and can change the way it is shown in this figure when the injection of proppant per stage in the completion design of hydraulic fracturing changes from 188,000 lbs. to 993,000 lbs. as long as all other variables that have been used to build this model will remain the same for this specific well.

If the proppant per stage is 188,000 lbs., then the Gas Production can increase from 0.1 million MSCF to 0.5 million MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft. If the proppant per stage increases to 680,000 lbs., then the Gas Production can increase from 0.15 million MSCF to 1.03 million MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft.

For Well #110 (the middle graph in Figure 18) the Type Curves explains that 6 Month Cumulative Gas Production can increases from 100,000 MSCF to 950,000 MSCF as the “Stimulated Lateral Length (ft.)” of this well increases from 3,500 ft. to 10,000 ft. This increase of Gas Production as a function of Stimulated Lateral Length is non-linear and can change the way it is shown in this figure when the proppant per stage changed from 188,000 lbs. to 993,000 lbs. as long as all other variables that have been used to build this model will remain the same for this specific well.

If the proppant per stage is 188,000 lbs., then the Gas Production can increase from 100,000 MSCF to 350,000 MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft. If the proppant per stage increases to 838,000 lbs., then the Gas Production can increase from 150,000 MSCF to 830,000 MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft.

For Well #066 (the bottom graph in Figure 18) the Type Curves explains that 6 Month Cumulative Gas Production can increases from 0.1 million MSCF to 1.22 million MSCF as the “Stimulated Lateral Length (ft.)” of this well increases from 3,500 ft. to 10,000 ft. This increase of Gas Production as a function of Stimulated Lateral Length is non-linear and can change the way it is shown in this figure when the proppant per stage changed from 188,000 lbs. to 993,000 lbs. as long as all other variables that have been used to build this model will remain the same for this specific well.

If the proppant per stage is 188,000 lbs., then the Gas Production can increase from 0.1 million MSCF to 0.6 million MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft. If the proppant per stage injection of the completion design increases to 993,000 lbs., then the Gas Production can increase from 0.25 million MSCF to 1.25 million MSCF as the Stimulated Lateral Length increases from 3,500 ft. to 10,000 ft.

Repeating what was mentioned before, Shale Analytics (as well as all the applications of AI and Machine Leaning in Petroleum Data Analytics) does not use any mathematical equations to generate data or to perform any kind of calculations to generate the types of type curves that are shown in Figure 17 and Figure 18. Actual field measurements that are shown in Figure 19 were the source of the Shale Predictive Analytics that generate the type curves demonstrated in Figure 17 and Figure 18. The complexity of the data shown in Figure 19 clarifies the quality of the Explainable Artificial Intelligence (XAI) that is used in Petroleum Data Analytics.

Explainable AI Model for Conventional Reservoirs – Top-Down Modeling

Type Curves shown in Figure 20, Figure 21, and Figure 22 were developed for a mature offshore field in northern Africa using Top-Down Modeling (TDM) that is a purely Data-Driven Reservoir Modeling[1] technology. These figures explain how and to what degree parameters such as Gas-Lift (Figure 20), Completion (Figure 21), and Porosity (Figure 22) can historically influence oil production for the entire field.

The well-behaved characteristics of these type curves that were developed using field measurements rather than mathematical equations clearly show how “eXplainable Artificial Intelligence (XAI)” has been used for the development of Top-Down Modeling.

As was mentioned before, Top-Down Modeling (AI-based reservoir simulation and modeling) does not use any mathematical equations to generate date or to perform any kind of calculations to generate type curves that are shown in Figure 20, Figure 21, and Figure 22. Actual field measurements that are shown in Figure 23 were the source of the Top-Down Modeling reservoir simulation that generates the type curves demonstrated in Figure 20, Figure 21, and Figure 22. The complexity of the data shown in Figure 23 clarifies the quality of the Explainable Artificial Intelligence (XAI) that is used in Petroleum Data Analytics.

[1] Book published by Society of Petroleum Engineers (SPE) – Data-Driven Reservoir Modeling. https://store.spe.org/DATA-DRIVEN-RESERVOIR-MODELING-P1054.ASPX