Decision making can be challenging when we are faced with interpreting extensive amounts of data . Machine learning simplifies this process, by allowing companies to review the data in a shorter amount of time, discover patterns that likely were not previously observable and determine the best course of action.
In the planning stages of a well, a drilling engineer has to design an optimal casing program, by picking among all : the casing set depths, hole sizes and casing sizes . Think about how simple this process would be if he/she had access of a map that shows all the historical wellbores grouped by their similarity on the mentioned properties . This similarity map is a product of clustering, an unsupervised ML technique which is becoming quite popular in Pro Well Plan platform.
Our latest clustering model is about Casing Similarity. The dataset, provided by our customer, describes 5 features that will be the focus of our model: the setting depths in both tvd and md, the diameter of the hole, length of casing and section types which are shown in the legend below.
Since we won't be working with labeled data, we are seeking constant evaluation from our domain experts. For testing purposes I have used a simple visualization to show the casings in wellbores (video below).
Let's jump back to the concept of similarity map .Something to keep in mind is that the similarity coordinates are not the same as the real coordinates of the wellbores, but a product of merging and reducing high dimension features. The shorter the distance between wellbores ,the more similar they are. You can explore the results further by coloring out the embedded data by different properties.
It is likely time to consider how your company could begin to harness these technologies to remain competitive and we are here to help you !