Evaluation of substation whole life cycle probabilistic carbon emissions based on data-driven AI and semi-invariant method

To accurately evaluate the whole life cycle carbon emissions of 220 kV substations under the coupling of multiple uncertain factors, a probabilistic carbon emission assessment method integrating data-driven artificial intelligence (AI) with the semi-invariant method is proposed. First, based on the life cycle assessment (LCA) method, the substation is divided into four stages: construction, building material production and transportation, operation, and demolition, and the deterministic carbon emission accounting is completed by the emission factor method. Then, an intelligent quantitative model for inventory data is constructed based on data-driven AI, which is combined with the semi-invariant method and Gram-Charlier series expansion to fit the probability distribution function of carbon emissions. Taking 10,000 Monte Carlo simulations (MCS) as the benchmark, a comparative verification is carried out with 6,000 Latin hypercube sampling (LHS). The results show that the relative width errors of the 95% confidence intervals of carbon emissions in each stage obtained by this method are all lower than those of LHS, with a significant improvement in calculation efficiency. The proposed framework integrates AI-based inventory processing, data-quality-driven uncertainty parameterization, and semi-invariant probabilistic propagation, thereby providing an efficient and reproducible tool for uncertainty-aware carbon-emission assessment of substations.