Use metaheuristics to improve the quality of drilling real-time data for advance artificial intelligent and machine learning modeling. Case study: Cleanse hook-load real-time data

The drilling engineers are overmild with huge amount of data-points, argue the need to develop Artificial Intelligent (AI) and Machine Learning (ML) models to crunch these huge amount of data generating decision-like information. There are a lot of challenges developing such approach, varying from computational power, lack of subject matter experts, and develop the optimum algorithm. But the main bottleneck is the quality of the data. Regardless of how advance AI/ML model, if the data is bad, the model will generate bad result; garbage-in garbage-out. The scope of this paper is to use metaheuristics models to improve the data quality. The process start by extracting Hook-Load drilling real-time data. And explore the raw data quality using visualization and statistical methods. Then apply several Metaheuristics models to generate functional approximation equation that identify/ follow the trend of the good-quality data. This will be by employ multiple scenarios with different degree of randomness that lead to the highest matching which generate the high quality level. The process will cover different technique including Greedy, Hill-Climbing, Random Search, and Simulated Annealing. During this process hundreds of thousands of scenarios will be conducted to simulate the Hook-Load data, to identify the optimum functional approximation equation that match the best data quality. Which can then safely integrated into the advance Artificial Intelligent and Machine Learning models. Running such process require an expensive computational cost, since it includes huge amount of real-time data need to be process under complex advance models. Moreover it require a deep understanding of the internal process of each models to ensure finest manipulating them to get the optimum data quality result. Running these scenarios, lead successfully to functional approximation that spill the data behavior, with Mean Absolute Error (MAE) equal to 10.5. It is worth height that functional approximation is very expensive in term of time and complexity, but it generate the highest quality result, leading to better AI/ML model. Moreover it is the most dynamic approach allowing it to be applied in other drilling real-time parameters as well. Utilizing Metaheuristics approach to improve the data quality is new to the upstream domain in general, with almost no application in drilling in specific. The novelty is to introduce this advance technique into the drilling real-time data domain, it will sharply improve the data quality leading to higher Artificial Intelligent and Machine Learning prediction/ analytical models. It worth mentioning that such approach will run all those simulation/ scenarios and adjust itself automatically with almost no manual interference. Leading to self-data-driven data-quality model.