Lightning nowcasting is crucial for ensuring safety and operational continuity in weather-exposed industries such as mining. This study evaluates three machine learning (ML)-based approaches for predicting lightning using dual-polarimetric weather radar data collected in the eastern Amazon, Brazil. The strategies propose advances in literature in three ways by involving (i) grouping radar variables by temperature layers, (ii) statistical summaries at key altitudes, and (iii) analyzing all the 18 levels of reflectivity data combined with Principal Component Analysis (PCA) dimensionality reduction and ensemble models. For each approach, models such as Random Forest, Support Vector Machines, and XGBoost were trained and tested using data from 2021–2022 with class balancing and feature engineering techniques. Among the approaches, the PCA-based ensemble achieved the best generalization (recall = 0.89, F1 = 0.77), while the layer-based method had the highest recall (0.97), and the altitude-based strategy offered a computationally efficient alternative with competitive results. These findings confirm the predictive value of radar-derived features and emphasize the role of feature representation in model performance. Additionally, the best model was integrated into the operational LEWAIS alert system, and four integration strategies were tested. The strategy that combined alerts from both ML and LEWAIS systems reduced the failure-to-warn rate to 0.0531 and increased the lead time to 10.18 min, making it ideal for safety-critical applications. Overall, the results show that ML models based solely on radar inputs can achieve robust lightning nowcasting, supporting both scientific advancement and industrial risk mitigation.
Lightning strikes are a major hazard in tropical regions, especially in northern Brazil, where open-area industries such as mining are highly exposed. This study proposes an octant-based multi-objective optimization approach for spatial lightning alert systems, focusing on minimizing both false alarm rate (FAR) and failure-to-warn (FTW). The method uses NSGA-III to optimize a configuration vector consisting of directional radii and alert thresholds, based solely on historical lightning location data. Experiments were conducted using four years of cloud-to-ground lightning data from a mining area in Pará, Brazil. Fifteen independent runs were executed, each with 96 individuals and up to 150 generations. The results showed a clear trade-off between FAR and FTW, with optimal solutions achieving up to 16% reduction in FAR and 50% reduction in FTW when compared to a quadrant-based baseline. The use of the hypervolume metric confirmed consistent convergence across runs. Sensitivity analysis revealed spatial patterns in optimal configurations, supporting the use of directional tuning. The proposed approach provides a flexible and interpretable model for risk-based alert strategies, compliant with safety regulations such as NBR 5419/2015 and NR-22. It offers a viable solution for automated alert generation in high-risk environments, especially where detailed meteorological data is unavailable.
1 January 2021 in Procedings Do Xv Simpósio Brasileiro de Automação Inteligente
Com o objetivo de impedir a entrada de objetos ilícitos que comprometam a segurança de funcionários, visitantes e detentos em instalações penitenciárias, a inspeção a partir de imagens radiológicas é amplamente utilizada para o monitoramento de ingressantes a complexos carcerários. Atualmente essa fiscalização é realizada por um operador que analisa imagens obtidas por equipamentos de raio-x, buscando identificar algum objeto suspeito. No entanto, tal processo é lento, dispendioso e repetitivo. Nesse sentido, o objetivo desse trabalho é desenvolver uma solução baseada em aprendizado de maquina não supervisionado, denominada Autoencoder Convolucional, cuja finalidade é auxiliar o operador na identificação de objetos ilícitos em imagens de raio-x de corpo inteiro. Para isso, foram coletadas e classificadas por um especialista 2.621 de imagens sem indicações de objetos suspeitos e 130 imagens de com suspeita de objetos ilícitos. Em seguida, um Autoencoder Convolucional foi treinado e avaliado, para detecção de anomalias. Os resultados obtidos indicam que a solução é promissora e pode ser utilizada no auxílio à tomada de decisão do operador do equipamento de raio-x.