Effluent Ammonia Nitrogen Prediction Based on Modular Neural Networks

To address the challenge of real-time and accurate detection of effluent ammonia nitrogen in urban wastewater treatment, this paper proposes a soft-sensing method for effluent ammonia nitrogen based on an improved density peak modular neural network. First, the improved density peak algorithm calculates sample density and distance to avoid noise becoming cluster centers, while adaptively determining the cluster centers and their number, thereby decomposing the prediction task and significantly enhancing the model's ability to handle complex problems. Second, a novel add-remove strategy is designed for different subsets, enabling the construction of compact sub-networks models based on sample characteristics, which improves the learning generalization and accuracy of the sub-networks. Finally, the effectiveness of the method is validated through benchmark experiments and applied to real wastewater treatment processes to predict effluent ammonia nitrogen concentrations, achieving accurate prediction results.