Statistical Benchmarking of Refinery Configuration Resilience: A Systems Engineering Approach to Feedstock Heterogeneity in the Nigerian Context

Abstract Nigeria is Africa's largest crude oil producer yet imports the majority of its refined products, a paradox the Petroleum Industry Act 2021 and the Dangote Refinery are designed to resolve. As domestic capacity expands, configuration design decisions determine whether new refineries can sustain throughput across the heterogeneous crude slate they will actually encounter. The Nelson Complexity Index (NCI) and Solomon Energy Intensity Index (EII), the dominant evaluation instruments, are static under feedstock variation and cannot predict how a refinery's outputs change when input quality shifts. This paper introduces the Resilience Coefficient (RC), derived from a multivariate OLS regression framework applied to the EIA Refinery Capacity dataset (EIA, 2025) comprising 116 U.S. refinery sites. The dependent variable is the System Flexibility Metric (SFM), the ratio of total conversion capacity to atmospheric distillation capacity. Three predictors are specified: Hydroprocessing Intensity Index (HPI), Vacuum-to-Atmospheric Ratio (VAR), and NCI. Critically, an HPI×VAR interaction term is introduced to capture the joint effect of deep hydroprocessing combined with vacuum distillation, the precise hardware combination required to process heavy, sour crudes. Model 2, the improved specification, achieves R2 = 0.993 (adjusted R2 = 0.993, F(5,110) = 3,191.26, p < 0.001) with RMSE of 0.026 SFM units. The interaction term HPI×VAR is negative and significant (β = −0.098, p = 0.0005), confirming that the joint presence of hydrotreating and vacuum distillation capacity, without accompanying conversion units, does not confer resilience. NCI remains the dominant positive predictor (β = 0.166, p < 0.001). Resilience Coefficients range from RC = −1.42 (Topping) to RC = +0.54 (Full-Conversion). Nigerian modular refineries, mapping to the Topping archetype, face a 1.96 standard-deviation RC deficit relative to full-conversion designs (Cohen's d = 3.02, p < 0.001), quantifying their stranded-asset exposure.