An Intelligent Selection Method of Main Controlling Factors for Tight Gas Reservoirs Productivity Based on Improved Harris Hawk Algorithm

Identifying the main controlling factors of oil and gas productivity and making accurate forecasts is crucial for efficient development and reservoir reconstruction. Tight gas reservoirs have complex geological conditions and high-dimensional, nonlinear factors that traditional methods struggle to analyze, complicating the identification of main factors and accurate productivity prediction. In the present work, an improved Harris hawk algorithm (TVLHHO), incorporating a nonlinear escape energy strategy and a time-varying leader structure, is proposed for the feature selection of the main controlling factors of tight gas productivity. The algorithm expands the search space of feature subsets, enhances convergence speed, reduces the risk of local optima, and ensures the accuracy of feature selection. Using a certain tight sandstone gas field as a case study, 59-dimensional features, including geological, logging, and fracturing properties, were used as inputs to study the main controlling factors affecting unimpeded flow rate. Initially, Pearson correlation analysis and XGBoost were used for preliminary feature selection, reducing the features to 23 dimensions. The TVLHHO algorithm was then employed to optimize the selection of the main controlling factors. Through an iterative process of updating the feature subsets and validating predictions, the optimal controlling factors identified included displacement fluid, deviation angle, azimuth angle, fracture half-length, gas relative density, perforation thickness, and initial gas saturation. The study shows that compared with six other well-known algorithms, TVLHHO not only demonstrates faster convergence but also achieves an R2 mean value exceeding 0.9 on the evaluator, resulting in the highest prediction accuracy. Furthermore, the main controlling factors selected by TVLHHO were used to predict the unimpeded flow rate, effectively identifying the distribution of high- and low-capacity wells. This validates the rationality of the TVLHHO feature selection results and demonstrates the algorithm's feasibility and effectiveness in practical applications. It provides a powerful tool for identifying the main controlling factors in tight gas reservoirs, addressing challenges related to high-dimensional data and complex relationships, and ultimately offering a more precise foundation for productivity prediction and optimization.