Browsing by Autor "Alejandra Tabares"

Now showing 1 - 7 of 7

A Hybrid Conv1D-LSTM Model with Temporal-Difference Reinforcement Learning for Error-Corrected Gas Forecasting in Critical Mining Environments
(2025) Sebastián Vivas; Pablo Andrés Gómez Monsalve; Alejandra Tabares
<title>Abstract</title> Gas accumulation in underground mining remains a critical safety concern, with hazardous levels of methane (CH4), carbon dioxide (CO2), and oxygen (O2) deficiency posing constant risk of accidents. Traditional forecasting approaches such as ARIMA, artificial neural networks (ANNs), and digital twins (DTs)—struggle to adapt in real time due to their reliance on static historical data and stationarity assumptions, leading to delayed or inaccurate anomaly detection under non-stationary conditions. In response to these limitations, we propose a hybrid forecasting architecture that integrates one-dimensional convolutional neural networks (Conv1D) and long short-term memory (LSTM) units with a reinforcement learning (RL) module based on semi-gradient temporal-difference learning. The proposed framework combines hierarchical feature extraction and temporal memory from the ANN with a correction mechanism via residual-driven reward optimization in RL. Evaluated on minute-resolution gas sensor data from a high-risk underground coke mine in Norte de Santander, Colombia, the hybrid model achieved substantial improvements in predictive accuracy. our model reduces CH4 RMSE from 4.56% with ARIMA to 0.91% with Conv1D–LSTM and further to 0.89% with RL, lowers MAE from 0.89% to 0.52% then to 0.47%, and cuts SMAPE from 82.9% to 74.2% and finally to 31.5%. For CO2, RMSE decreases from 0.21% to 0.17% and to 0.03%, MAE from 0.15% to 0.12% and to 0.02%, and SMAPE from 70.8% to 56.6% and to 8.4%. For O2, RMSE falls from 0.59% to 0.29% and to 0.09%, MAE from 0.26% to 0.21% and to 0.06%, and SMAPE from 1.34% to 0.99% and to 0.23%. These comprehensive gains demonstrate RL’s effectiveness as an error‑correction layer, enhancing both accuracy and responsiveness of deep sequence models in volatile, safety‑critical mining environments. These findings are a solid foundation for future studies in another high-risk environments.
Advancing Mg Energy Management: A Rolling Horizon Optimization Framework for Three-Phase Unbalanced Networks Integrating Convex Formulations
(RELX Group (Netherlands), 2025) Pablo Cortés; Alejandra Tabares; John F. Franco
Ai-Powered Dynamic Fault Detection and Performance Assessment in Photovoltaic Systems
(RELX Group (Netherlands), 2024) Nelson Salazar-Peña; Alejandra Tabares; Andrés González-Mancera
Ai-Powered Dynamic Fault Detection and Performance Assessment in Photovoltaic Systems
(RELX Group (Netherlands), 2025) Nelson Salazar-Peña; Alejandra Tabares; Andrés González-Mancera
Harnessing Implicit Cooperation: A Multi-Agent Reinforcement Learning Approach Towards Decentralized Local Energy Markets
(RELX Group (Netherlands), 2026) Nelson Salazar-Peña; Alejandra Tabares; Andrés González-Mancera
MARLEM: A Multi-Agent Reinforcement Learning Simulation Framework for Implicit Cooperation in Decentralized Local Energy Markets
(RELX Group (Netherlands), 2025) Nelson Salazar-Peña; Alejandra Tabares; Andrés González-Mancera
Optimal Hybrid Backup Systems for Substation Auxiliary Services During Outages Using Stochastic Programming
(RELX Group (Netherlands), 2024) Matheus Holzbach; Norberto Abrante Martinez; Lucas Teles Faria; Alejandra Tabares; José F. Resende; John F. Franco