eproducibility Package for Topology-Constrained Multi-Type Hawkes Modeling of Fault Cascades in the CEET Lomé Distribution Grid

This reproducibility package contains the final code, analytical inputs, model outputs, and validation materials supporting a topology-constrained multi-type Hawkes analysis of fault cascades in the electricity distribution grid operated by Compagnie Énergie Électrique du Togo (CEET) in Lomé, Togo. The study is based on 7,152 classified fault events recorded from 2017 to 2023 across 77 active feeders. A documented physical network comprising 51 feeders and 94 electrical corridors is used to represent cross-feeder propagation, while the remaining 26 feeders retain local self-excitation without artificially assigned interconnections. The package includes the final implementations and results for: Poisson, same-feeder, uniform cross-feeder, and topology-constrained Hawkes models; refined temporal-decay parameter selection; feeder-topology and adjacency-normalization analyses; event-taxonomy sensitivity analyses; nested rolling-origin validation; monthly block bootstrap of held-out likelihood differences; timestamp consistency, simultaneous-event, and August 2023 recording-gap audits; source-consistent NOAA/NASA weather reconstruction for 2017–2023; nested weather robustness analysis using a shared multiplicative baseline modifier across fault types. Only the latest authoritative scripts, required source inputs, canonical processed datasets, and final analytical outputs are included. Superseded, diagnostic, erroneous intermediate, backup, and temporary files have been excluded. The primary scientific model is the topology-constrained multi-type Hawkes model without weather covariates. Weather-adjusted models are included as robustness analyses. Hawkes excitation, endogeneity measures, and weather coefficients are interpreted as predictive associations and not as evidence of physical causality.