Research on the relationship between electrical conductivity and uranium concentration in sandstone-hosted uranium...

During in-situ leaching (ISL) of sandstone-hosted uranium deposits, electrical conductivity (EC) is strongly coupled with leachate uranium concentration (U). As a key monitoring indicator reflecting solution ionic strength and seepage-channel development, EC evolution is closely associated with uranium leaching performance and can support operational decisions (e.g., injection recipe and injection-production scheduling), thereby improving process stability and efficiency. However, influenced by hydrochemical parameters such as sulfate (SO₄^2-) concentration, the EC-U relationship exhibits pronounced nonlinearity and stage-dependent behavior that is difficult to capture using conventional empirical models. To address this challenge, a deep-learning-based hybrid model, termed CA-CNN-BiLSTM-LightGBM (LGBM), has been developed for uranium concentration prediction and EC-U response characterization. The CNN extracts local coupled features from EC and related hydrochemical variables, the BiLSTM captures the lagged U response to EC variations and the characteristic rise-plateau-decline pattern, and channel attention with an autoencoder reconstruction constraint enhances feature weighting and representation stability. LightGBM is further introduced to correct residuals of the deep backbone, improving fitting accuracy under extreme operating conditions. Experiments demonstrate that the proposed model outperforms Ridge, LightGBM, LSTM and GRU baselines, achieving an R² of 0.962, MAE of 1.121 mg/L and RMSE of 1.828 mg/L on the test set, providing a reliable data-driven tool for process parameter optimization and intelligent control in sandstone-hosted uranium ISL.