This study aims to predict environmental pollution risk associated with liquid effluents from a mineral analysis laboratory located in Kolwezi, Lualaba Province (LP), Democratic Republic of Congo (DRC). The analysis focuses on the temporal variability of trace metal concentrations, such as arsenic (As), copper (Cu), iron (Fe), zinc (Zn), lead (Pb), and nickel (Ni), over a four-year monitoring period from January 2022 to January 2026. A hybrid Long Short-Term Memory (LSTM) neural network incorporating an attention mechanism was implemented in Python to model and forecast the behavior of these contaminants. Using this model, the data were carefully preprocessed to ensure their quality, and the model's performance was measured using Mean Squared Error (MSE) and the coefficient of determination (R²). The measured concentrations were compared against the DRC Mining Regulation discharge limits (Decree No. 038/2003, as amended by Decree No. 18/024 of June 2018). Historical monitoring revealed that copper (max 35.907 mg/L in February 2022) and iron (max 14.140 mg/L in January 2022) exhibited the most pronounced exceedances of regulatory thresholds, though concentrations decreased substantially over time. The LSTM-Attention-LSTM model demonstrated excellent predictive performance, with R² values exceeding 0.99 for all six elements and MSE values ranging from 3.1 × 10⁻¹⁰ (As) to 0.9052 mg²/l² (Cu). One-year forecasts (January 2026 to January 2027) indicate that arsenic, lead, nickel, and iron are expected to remain below regulatory thresholds, while zinc is forecast to approach 2.7 mg/L by mid-2026 and copper may fluctuate near its limit. These findings underscore the urgent need for systematic pre-discharge treatment at the laboratory and demonstrate the potential of deep learning models as early-warning tools for environmental compliance monitoring in mining-adjacent analytical facilities.
Musala et al. (Fri,) studied this question.