Prediction of Base Parameters for Production Estimation of Greenfield Deposits

About DotGEO FastCalc

In DotGEO I am working for system called FastCalc – the aim of the system is provide a report about possible accumulated debit of oil, liquid, initial daily debits and many other parameters that are needed for development plan.

Being almost 10 years in industry, especially such conservative as oil and gas, the hardest part is to make specialists believe that your product is approved and they really can understand and affect on calculations – no black-boxes allowed. So, here and after I will disclose some base ideas of FastCalc and its logic.

While developing idea of software I wanted to follow several simple rules:

• Every calculation step must be clear and approved and used by society of engineers – Arps curves, Unit economy and others
• AI must be used to fill the steps that require human – so in my case these steps are calculations of input parameter for further calculations
• There is no need for some “average solution” – the market already has a number of great software from #slb #bakerhughes #halliburton and other giants of industry, so probably my target audience is focused on those who need fast and simple result right now, instead of long and pedantic

It seems that logic that I placed in FastCalc corresponds these rules and internal pipeline looks like this:

Let`s talk now about the one blue point on the pipeline – it is a simple ML algorithm that finds initial daily debit of new wells basing on petrophysical properties.

Small dicusiion about ML approach in Oil&Gas

We all understand that dealing with ML the key parameter is accuracy of a model. The accuracy itself depends on training dataset and quality of input values. Input values for the model are petrophysical parameters that can be estimated not precise enough, especially if the deposit is young. Also, there may be not enough measured parameters – for example just couple of them. Well, in that case we can`t hope that results of whole prediction will be good enough, but anyway that’s not a problem for FasCalc.

If you ever had an experience in training models, you probably remember the training graph with y-axis for error, and x-axis for amount of training epochs. In our case the most interesting visualisation is a graph with error vs number of features. FastCalc is build with self-training logic, so every user could build prediction based on his company relevant experience, so amount of features given for training and for prediction is more crucial that amount epochs.

My training dataset has list of 24 features – the most important from geological point of view are: porosity, permeability, pressure, viscosity of oil and water, oil saturation. The others are more specific – normally you will have them after first exploration well.

Here is a visualisation of how amount of features impacts the accuracy of ML prediction of extraction coefficient and initial oil debit:

Here we can see that even small additional number of features drives to significant improvement of prediction, but since the amount features is more than 5-6, the accuracy stays almost the same. Seems irrational, but let`s keep in mind that we are working with geology, where everything is described empirically – and here is a reason why. Let`s build correlation matrix for my dataset and try to analyse it:

As you can see on the last image (mask of correlation matrix, with hidden values in range of (-0.55, 0.55), every of 24 parameters has at least one other with good correlation, and just couple of unique parameters that has no correlation with others. What does it mean? It means that adding first several parameters for training and applying the model for assessment of needed geological parameter is crucial, and even it may seems a trick, one can easily get good quality of prediction with just several measured or guessed parameters. In FastCalc I am focusing on building a software that helps people to get a result in such cases, where others will say – huh, not enough data.

Try FastCalc

I invite you to experience FastCalc. Contact me for a trial version of DotGEO FastCalc and see how it can transform your prediction process.

#AIinOilAndGas #MachineLearning #OilAndGasIndustry #Petrophysics #TechnologyInnovation #DataDrivenDecisions #FastCalc #DotGEO #EnergyTech #FutureOfEnergy #Geology #Engineering