Advanced Well Planning Using Natural Language Processing NLP and DataScience Models:Maximizing the Value of Data to Mitigate Costs and Risks inNew Wells

摘要

A new methodology using data science models and natural language processing(NLP)model has beendeveloped,enabling rapid well screening and automated recommendations of key planning parameters.Well Planning and Drilling engineers are empowered by this approach,able to perform quick analysis ofall available relevant data,to come up with drilling plans and budgets,and a template for new wells basedon the entire field drilling experience."Analog"well and"Cluster"were introduced as key concepts in the methodology.Analog wells are wellsthat have similar characteristics and are completed in the same reservoir as the proposed new well.Clusteringtechniques used were tested on a publicly available data set,shared by Equinor for the Volve Field.Within the given data set,5 clusters were identified by the similarity-based clustering algorithm.Foreach well cluster,a distance was calculated between wellbores and the averaged features of that cluster.Wellbores in the same cluster were used to anticipate type,duration and impact of NPT events.The Similarity Index was introduced and a representative well bore for each of the clusters was devised.This representative wellbore was used as a benchmark for recommending casing,ROP,mud density,expected pore pressure,costs,drilling and completion time,expected NPT,among other parameters.The results suggested that using a similarity index rather than simply using the nearby wells identifiedstronger similarities between wells than would be found based on proximity.It was determined that in thisdata set there were a number of wells that were not in the close proximity to the proposed well which werein fact closer analog wells than those which would have been used by taking the nearby wells.The pre-calculation of clusters and similarity indices made the process of initial estimation for new wellsvery rapid and bias-free.For the NLP exercise,the accuracy of classifying NPT achieved was 98% through selecting the optimaltechnique.The approach also highlighted the need for a dictionary with common abbreviations used in thedrilling industry,to standardize and share meanings of terms across companies and even teams.The proposed method overcomes the major limitations of user-biased process of manually gathering,with a limited data set transformed to an objective process,is inclusive of all wells in the field,and thevalue of automating results.The technique is not a substitute for the detailed well engineering design required to develop a well designto AFE approval standard,however,the tool enables a much needed acceleration in well design achievingsignificant reduction in time,costs and risks.