Development of Time Series Drilling Datasets for Stuck Pipe Prediction Using Volve Field Data

Abstract Numerous machine learning algorithms are applied in the oil and gas industry. However, the data used in these studies are difficult to obtain due to various limitations. Due to the lack of benchmark datasets, it is challenging to make performance comparisons across different algorithms. Volve field dataset was made public by Equinor, which provides raw data availability for the development of drilling and completion datasets. In this paper, we utilize the time-based drilling data from the Volve drilling platform and transform it into time series datasets for stuck pipe prediction. Specifically, we introduce our concepts and principles for data development, the rules for selecting time intervals and attributes, the challenges encountered during the data development process and the methods for overcoming them. We discuss the applicability of these methods, the issues they bring, and their impact on data quality. Furthermore, we provide a well development case that includes complex data. The results indicated that our research shows promise in providing time series reference datasets for the application of machine learning algorithms in stuck pipe prediction. We aim to provide a reference methodology for the development of raw data, reducing barriers to data utilization. We hope to provide data availability, possibly even serving as a reference benchmark dataset for the further applications of machine learning algorithms in the oil and gas sector. Our datasets are made publicly available on GitHub : https://github.com/promiseeee/Time-series-stuck-pipe-prediction.