A Method for Filling the Missing Values of Pipeline Data Based on Random Forest and BP Neural Network

The dataset of oil pipelines often contains missing values, which can impact leak detection. This paper proposes a novel algorithm for imputing missing values in pipeline data, called RFKMI. The algorithm starts with preliminary imputation using the k-Nearest Neighbors (KNN) method. Next, K-means clustering stratifies the data for more precise imputation within each cluster. Missing values are then imputed using two distinct models: a parameter-optimized Random Forest model and a BP Neural Network model. An adaptive fusion algorithm combines two input datasets into a single dataset and then trains a classifier to identify pipeline leakage data. Experiments on nine UCI datasets show that RFKMI effectively imputes missing data across various missing rates. Test on simulated oil pipeline data indicates high classification accuracy, suggesting improved reliability in oil pipeline monitoring systems.