Source apportionment of pollution in the Tianjin Haihe river sluice based on the TCN-APCS-MLR model

River pollution identification in complex watersheds like the Haihe River Sluice faces challenges due to temporal dynamics and nonlinear driving mechanisms. Traditional receptor models (CMB, PMF, APCS-MLR) often lack the sensitivity required for accurate, timely source tracing. This study proposes the TCN-APCS-MLR model, integrating TCN with Absolute Principal Component Score-Multiple Linear Regression.Using daily water quality data from 2022 to 2024, the model leverages TCN's causal and dilated convolutions to extract temporal dependencies, followed by PCA for dimensionality reduction and MLR for quantitative source estimation.This approach enhances temporal feature modeling while preserving interpretability.Results indicate:Superior Performance: The TCN-APCS-MLR model outperformed traditional methods, with R² values for EC, CODMn, and TN exceeding 0.85. Error Reduction: RMSE and MAE decreased by 20%-30% compared to the standard APCS-MLR model.Feature Extraction: The cumulative variance contribution of the first four principal components reached 81.87% (an 8% improvement).Source Apportionment Results:The Haihe River Sluice is primarily influenced by:Urban domestic and industrial wastewater (26.5%);Ecological self-purification processes (26.6%);Natural hydrological disturbances (25.1%);Agricultural non-point source pollution (21.8%).By incorporating deep learning, this model effectively addresses the limitations of linear models in complex estuarine environments, providing a robust technical framework for dynamic watershed management and ecological restoration.