Dynamic evaluation of drilling leakage risk based on fuzzy theory and PSO-SVR algorithm

Abstract

In recent years, artificial intelligence has gradually penetrated into various fields, and has become a research hotspot. The modern industrial upgrades and transformation of the petroleum industry, makes it closer to the direction of intelligence. For the research of drilling risk evaluation, choosing the right evaluation model to achieve real-time risk dynamic evaluation which is important for risk judgment and response time. However, drilling system never considered as a complex system in the research of drilling risk assessment. When the sensor of the well site collects the relevant parameters, the remote monitoring system carries on the real-time data analysis, because of the instrument or transmission process, the drilling parameters appear fuzziness and randomness. To realize real time dynamic evaluation of drilling risk this paper proposed a fuzzy multilevel algorithm based on Particle swarm optimization (PSO) to optimize Support vector regression machine(SVR), and takes drilling leakage risk as an example. And two main objectives has been achieved. The first is to establish a fuzzy multi-level drilling leak risk evaluation system. The second is to use the PSO-SVR algorithm to study the risk evaluation results and realize the real-time dynamic risk evaluation. This paper first summarizes the characterization phenomena and laws of the occurrence of acquisition and loss parameters, and uses this as an indicator to establish a multi-level index system for risk assessment. Second, combined with fuzzy theory, a risk assessment model is established. And in final, the parameters C and g of the SVR model are optimized by using the SVR algorithm improved by PSO, which solves the problem that the parameters such as penalty factor c, kernel function k and sensitivity coefficient ε are difficult to select in the traditional SVR model, improves the accuracy of the model, and realizes more accurate real-time dynamic evaluation of risk. The algorithm proposed in this paper achieves two goals. Taking the XX oilfield as an engineering example, the results show that the accuracy of the PSO-SVR model can reach 99.99%, with high convergence degree, which is obviously higher than that of the multilayer perceptron neural network model.