Protsiuk V. Neural network system for automated prediction of the dynamics of controlled parameters of the oil and gas well drilling process

The aim of the dissertation research is to develop and implement an intelligent system that uses neural networks to automatically predict changes in controlled drilling parameters. Such a system can significantly increase drilling efficiency, reduce the risk of accidents, and optimize resource consumption through more accurate predictions and promt responses to changes in the drilling process. The thesis presents a theoretical generalization and solution of an actual scientific and practical problem of creating a neural network system for automating the prediction of the dynamics of the control parameters of the process of drilling oil and gas wells. This involves the development of methods and algorithms that allow anticipation of changes in drilling parameters with high accuracy, thereby ensuring the efficiency and safety of the drilling process, as well as optimizing costs and resources. Solving this problem is important for the oil and gas industry of Ukraine, as it makes it possible to optimize the technical and economic indicators of the oil and gas well drilling process. The scientific novelty of the dissertation is determined by the following points, that describe what was achieved for the first time and what was improved:  the development of methods and models that allow to effectively eliminate and compensate for the impact of anomalies in the data of the oil and gas well drilling process, which is achieved through the use of neurofuzzy technologies that can work in conditions of incomplete or fuzzy information, as well as complex nonlinear functional dependencies;  the construction of a neuro-fuzzy model for assessing the impact of anomalies, which includes a fuzzy product knowledge base, mechanism of fuzzy logical conclusion and a model tuning system in the form of the proposed neural network optimization methodology using gradient descent and regularization methods, which avoids overtraining and ensures efficient operation of the model in the extended drilling data space.  improved the methodology of using the Hamming distance instead of the multiplicative aggregation function for calculating confidence levels, which allows better work with quantitative sensor data.  improved the formal analytical evaluation of anomalies in oil and gas well drilling data; a general model of the drilling oil and gas wells process control system that uses mathematical formalization to describe and control drilling processes with regard to anomalies; models based on dynamic systems that allow to formally describe control processes as a spatio-temporal system with appropriate coordinates; algorithms for controlling the drilling process, including estimates of process states depending on operating parameters and levels of anomalies; mathematical models that take into account the influence of anomalies on the initial parameters of the process.  further development of the anomaly detection algorithms based on rules with confidence factors that allow processing sensor data with high accuracy, taking into account various types of uncertainty, such as incompleteness, ignorance, blur, imprecision and ambiguity, which makes them more effective than traditional methods such as Gaussian algorithms or binary and fuzzy association rules. The practical value of the study, which includes the development of models for analytical evaluation of anomalies in oil and gas drilling data, is significant for the oil and gas industry. The main aspects of practical value are to improve the accuracy and reliability of drilling processes. As a result, the proposed research contributes to increasing the economic efficiency of drilling operations, reducing costs and risks, as well as improving the quality and safety of operations in the oil and gas industry. The results of the dissertation research were used in the real conditions of LLC “NADRA ENGINEERING” to optimize and improve the processes of drilling of oil and gas wells, increase work efficiency and reduce costs. Also, the results of the dissertation research were implemented in the educational process during the study of the discipline “Modern Methods of Working with Big Data” by students within the specialty 121 “Software Engineering”. The facts of implementation are confirmed by relevant documentation shown in the appendices of the dissertation. Keywords: automation of the drilling process, mathematical models, knowledge-based decision support systems, expert systems, forecasting methods, deep learning, machine learning, anomalies in data, anomaly detection algorithms, multilayer neural networks, uncertainty, fuzziness, confidence coefficients.