A New Well Loss Prediction Model was Created by Applying Genetic Algorithms to Enhance the RBF Neural Network

In the process of drilling operation, various types of complex drilling accidents will occur, of which the lost circulation accident is the most common and one of the most serious drilling accidents. For the sake of drilling safety and cost, people always want to predict or early warn drilling lost circulation accidents in advance. In recent years, due to the development of computer intelligence technology and integrated logging technology, while major oil companies vigorously implement digital and intelligent exploration and development, and radial basis function (RBF) neural network structure is simpler than the BP neural network, with global convergence, fast convergence, fast learning and excellent approximation. The SVM algorithm can be easily applied to extensive training datasets, hence the prevalent adoption of the RBF neural network approach. However, data analysis reveals limited reports on predicting drilling lost circulation with the RBF neural network algorithm. This paper aims to develop a predictive model for lost circulation during drilling, utilizing the RBF neural network algorithm. It will consider various factors, including geomechanical data, hydraulic parameters, the characteristics of drilling fluids, wellbore conditions, and the physical properties of the reservoir throughout the drilling operation. The genetic algorithm (GA) serves as a tool to fine-tune the parameters of the radial basis function (RBF) neural network. This optimization enhances the RBF neural network’s efficiency, facilitating quicker calculations, simplifying operations, and boosting predictive accuracy. Ultimately, it enables timely and accurate alerts for lost circulation, which holds significant practical implications for guiding engineering practices. Results from field experiments reveal that the RBF neural network, optimized through GA, achieves an average prediction error of 6.52%. In comparison, the unoptimized RBF neural network records an average error of 14.98%, the BP neural network shows an error of 17.49%, the SVM (support vector machine) algorithm has an average prediction error of 12.59%, and the SVR (support vector machine regression) method utilizing a Gaussian kernel function reaches an average error of 10.09%. These findings indicate that the newly developed model demonstrates strong adaptability for predicting lost circulation and has the potential for broader application.