Mechanism-grounded Bayesian anomaly inference for FPSO production systems under incomplete probabilistic information
Description
This dataset contains the numerical data used to support the offline probabilistic parameter optimization results reported in the manuscript “Mechanism-grounded Bayesian anomaly inference for FPSO production systems under incomplete probabilistic information”. The case study is based on a three-phase separator in an FPSO production system and uses a mechanism-grounded layered Bayesian network with 22 nodes and 31 directed propagation relationships. The dataset includes the principal numerical parameters used in the optimization, the convergence histories of the unified loss under different settings, the effects of the probability propagation tolerance ξ and posterior probability threshold τ, and the decomposition of the unified loss into probability propagation and posterior consistency components. Conditional-probability intervals are defined according to the computational implementation as the initial probability ±0.2 within the prescribed probability bounds. The dataset also records the fixed ξ and τ values examined in the comparative experiments and the final jointly optimized results. The figure data correspond to Figs. 8–10 of the manuscript and are provided mainly in CSV format, with one XLSX file for the unified-loss decomposition. The data can be processed using Python, MATLAB, Microsoft Excel, Origin, R, or other numerical data-processing software. The dataset is provided to improve the reproducibility, transparency, and independent verification of the numerical optimization results presented in the manuscript.
Files
Steps to reproduce
Steps to reproduce: Download Simulation_parameters.csv and all files in the Figure_data folder. Construct the Bayesian network for the FPSO three-phase separator according to the network topology described in the related manuscript. The case-study network contains 22 nodes and 31 directed propagation relationships. Initialize the probabilistic parameters using the values and parameter ranges provided in Simulation_parameters.csv. For each conditional probability, define the admissible interval as the initial probability plus or minus 0.2, while keeping the probability within its prescribed numerical bounds. Evaluate candidate probabilistic parameters using the unified loss described in the manuscript. The unified loss consists of probability propagation loss and posterior consistency loss. Apply Levy-flight particle swarm optimization to minimize the unified loss. The maximum number of optimization iterations is 500. Reproduce the seven experimental groups used for Figure 8. Experiment a excludes probability propagation consistency. Experiments b to f use fixed probability propagation tolerances of 0.1, 0.2, 0.3, 0.4, and 0.5, respectively. For each group, posterior probability thresholds of 0.3, 0.4, 0.5, 0.6, and 0.7 are tested and compared with continuous threshold optimization. Experiment g jointly optimizes the probability propagation tolerance, posterior probability threshold, and probabilistic parameters. Use the Figure 8 data files to reproduce the convergence curves of the unified loss under the different experimental settings. Use the Figure 9 data file to reproduce the comparison of the effects of the probability propagation tolerance and posterior probability threshold. Use the three Figure 10 data files to reproduce the decomposition of the unified loss under representative settings. For the joint optimization case, the expected final probability propagation tolerance is approximately 0.4335, the posterior probability threshold is approximately 0.3146, and the final unified loss is approximately 0.4335. The final solution contains no modeled probability propagation violations and no posterior inconsistencies. The provided files can be used to reproduce Figures 8 to 10 and verify the reported numerical optimization results. The complete source code is not included in the dataset.
Institutions
- China University of Petroleum, BeijingBeijing, Beijing
Categories
Funders
- National Science and Technology Major Project of ChinaGrant ID: 2026ZD1610000