A Novel Data Mining Approach in Preventing Casing Damage of Oil Production Wells

Because casing repair is expensive and the repair technology is not mature enough. Therefore, it is necessary to identify the casing damage risk well in advance and formulate effective casing damage prevention measures. In this paper, XGBoost and LightGBM machine learning algorithms are used to establish the casing damage prediction model for oil production wells. Taking the factors of dynamic production pressure difference and other factors leading to the casing damage of the production layer as input parameters and casing damage as the output parameter, a data set of 653 samples is established. The data set is split into a training set (80%) and hold-out set (20%).In the hold-out set, the casings of 111 production layers were not damaged. The hold-out set contains data corresponding to 131 production layers. Among them, 17 production layer casings were damaged, and the model predicted 13. The other 114 without casing damage are predicted correctly. The overall prediction accuracy is 96.9%. The main controlling factor of casing damage is production pressure difference. The casing damage prediction model is used to analyze the parameter sensitivity, optimize operation parameters such as production pressure difference, and reduce the casing damage probability of the oil production well.