Natural Gas compressibility factor and associated ML-based codes (ANN +GMDH+GP)

This repository provides a curated databank of dry-gas compressibility factor (Z-factor) measurements compiled from multiple literature sources, alongside the code and trained models used in our head-to-head evaluation of modern machine-learning approaches versus classical equations of state and correlations. The dataset contains 1,079 data points with pressures and temperatures harmonised to MPa and K. Pseudo-reduced variables were recomputed consistently (Ppr = P/Pc, Tpr = T/Tc) and Z was obtained/reconciled using standard industry practice. The data span approximately 0.2 ≤ Ppr ≤ 30 and 1.05 ≤ Tpr ≤ 3. Model development and validation in the accompanying code were conducted within the subset 0.162 < Ppr < 25.821 and 1.357 < Tpr < 2.42. Included scripts implement three predictive models—ANN with Levenberg–Marquardt (ANN-LM), Group Method of Data Handling (GMDH), and Genetic Programming (GP)—together with example outputs, weights/biases, and tabulated results to facilitate replication and extension. Users are encouraged to adapt and improve these models for their own dry-gas systems. Contents Data: Dry-gas Z-factor databank (CSV/XLSX), with P, T, Pc, Tc, Ppr, Tpr, and Z. Code: developed_best_ANNLM.m (ANN-LM model and training script) GMDH_DevelopedModel.m (GMDH model and sensitivity options) Developed GP model (GP expression / notes) Results: All models results_Weights and Biased of ANN.xlsx (trained weights/biases and model performance tables How to use Open the scripts in MATLAB (R2020b or later recommended). Load the provided data file; paths are relative to the repository root. Run each script to reproduce training, testing, and the reported metrics; adjust hyperparameters as needed. For deployment, apply models only within the validated range (0.162 < Ppr < 25.821; 1.357 < Tpr < 2.42). Potential applications Rapid Z-factor estimation for dry-gas reservoirs Benchmarking new ML models against published baselines Sensitivity analysis and uncertainty quantification Teaching materials for data-driven thermophysical modelling