城市化进程加速导致水污染问题日益严峻，其中出水氨氮浓度的实时精准监测是水质安全管理的核心挑战。传统物理化学检测方法存在滞后性高、成本昂贵等问题，难以满足动态水质管控需求。针对这一问题，本文提出一种基于模块化神经网络的氨氮软测量方法，实现出水氨氮实时、低成本、高精度预测。本文的主要研究内容如下：

（1）针对现有基于聚类的任务分解方法主要局限于聚类中心的在线更新而难以有效捕捉样本间潜在相似性特征的问题，提出了一种基于动态密度峰值聚类的在线任务分解算法。该算法将样本相似性度量融入传统的局部密度最大化准则和类间距离最大化准则，通过双重准则严格定义了簇中心的判定条件，随后基于动态相似性分析实现全局样本密度的实时更新和簇中心的自适应调整，从而实现任务层的动态划分。通过对两个基准任务和一个实际问题进行建模，算法实现了任务动态自适应调整，实验结果表明，与其他对比模型相比，该模型子网络复杂度更低，且模型预测性能更优与泛化能力更强。

（2）针对出水氨氮浓度变化具有时变性和非线性导致软测量模型建模精度差的问题，提出了一种在线自组织模块化神经网络的出水氨氮软测量模型。该模型引入在线任务分解算法，动态划分样本空间，有效缓解数据分布偏移引起的全局性能降低，提升了模型的动态跟踪能力。同时，采用自组织算法构建子网络，并提出双模态更新策略，分别对结构和参数进行动态调整，在降低模型复杂度的同时，兼顾结构灵活性与参数的收敛效率，增强了模型对复杂数据变化的适应能力。通过Mackey-Glass时间序列预测，与其他模型相比，该模型预测性能明显更强、复杂度更低。通过对出水氨氮进行软测量建模，与其他模型比较，模型性能和子网络数目至少提升64%、67%。

（3）针对传统氨氮测量方式的局限性。设计开发了一套基于MATLAB GUI的出水氨氮浓度软测量软件，软件集成了所提出的理论模型，围绕用户管理和出水氨氮浓度软测量两大核心功能，实现了数据可视化和快速预测分析，为污水处理过程出水氨氮软测量提供了高效的决策支持工具，助理水质管理的科学化和精准化。

With the accelerating process of urbanization, water pollution has become increasingly severe, and the real-time and accurate monitoring of effluent ammonia nitrogen concentration has become a core challenge in water quality management. Traditional physicochemical detection methods suffer from high latency and high costs, making them inadequate for dynamic water quality control needs. To address this issue, this paper proposes a modular neural network-based soft sensing method for ammonia nitrogen, enabling real-time, low-cost, and high-precision prediction of effluent ammonia nitrogen concentration. The main research contents of this paper are as follows:

(1) To address the limitations of existing clustering-based task decomposition methods, which primarily focus on online updates of cluster centers and struggle to effectively capture the underlying similarity features between samples, a novel online task decomposition algorithm based on dynamic density peak clustering is proposed. This algorithm integrates sample similarity measurement into the traditional criteria of local density maximization and inter-class distance maximization, establishing strict conditions for identifying cluster centers through a dual-criterion approach. Subsequently, a dynamic similarity analysis is used to achieve real-time global sample density updates and adaptive cluster center adjustment, thereby enabling dynamic partitioning of the task layer. By modeling two benchmark tasks and one real-world problem, the algorithm achieves dynamic and adaptive task adjustment. Experimental results demonstrate that, compared to other models, this method yields lower sub-network complexity, superior prediction performance, and stronger generalization ability.

(2) To address the problem of poor modeling accuracy in soft sensing models caused by the time-varying and nonlinear characteristics of effluent ammonia nitrogen concentration, an online self-organizing modular neural network-based soft sensing model is proposed. This model incorporates an online task decomposition algorithm to dynamically partition the sample space, effectively mitigating performance degradation caused by data distribution shifts and enhancing the model's dynamic tracking capability. Additionally, a self-organizing algorithm is used to construct subnetworks, and a dual-mode update strategy is introduced to dynamically adjust both the network structure and parameters. This approach reduces model complexity while maintaining structural flexibility and parameter convergence efficiency, thereby improving the model's adaptability to complex data variations. Through Mackey-Glass time series prediction, the proposed model demonstrates significantly better predictive performance and lower complexity compared to other models. In the soft sensing modeling of effluent ammonia nitrogen, the model outperforms others, with improvements of at least 64% in performance and 67% in the number of subnetworks.

(3) To address the limitations of traditional ammonia nitrogen measurement methods, a MATLAB GUI-based effluent ammonia nitrogen concentration soft measurement software has been designed and developed. The software integrates the proposed theoretical model and focuses on two core functions: user management and effluent ammonia nitrogen concentration soft measurement. It enables data visualization and rapid predictive analysis, providing an efficient decision-support tool for soft measurement in wastewater treatment processes. This enhances the scientific and precise management of water quality.

[1] Yin F, Xu C. Quantifying the inter-and intra-annual variations in regional water consumption and scarcity incorporating water quantity and quality[J]. Water Resources Management, 2020, 34: 2313-2327.

[2] Zhang X, Chen X, Xiao J, et al. Comparative study of different sewage sludge incineration treatments based on environmental and economic life cycle assessment[J]. Waste Management & Research, 2024, 42(5): 418-429.

[3] Chen Y, Cen G, Hong C, et al. A metric model on identifying the national water scarcity management ability[J]. Water resources management, 2018, 32(2): 599-617.

[4] 吴一平, 宋燕妮, 张广创, 等. 流域面源污染研究进展与展望[J]. 农业环境科学学报, 2024, 43(04): 711-720.

[5] Liu Z, Zhang C, Li Q. Monitoring of pollution factors of solidified body leachate in a drilling well site and its influence on the surface water environment[J]. Water Supply, 2021, 21(3): 1081-1089.

[6] Samkhaniani M, Moghaddam S S, Mesghali H, et al. A machine learning approach to feature selection and uncertainty analysis for biogas production in wastewater treatment plants[J]. Waste Management, 2025, 197: 14-24.

[7] Matveev I B, Serbin S I, Washchilenko N V. New combined-cycle gas turbine system for plasma-assisted disposal of sewage sludge[J]. IEEE Transactions on Plasma Science, 2017, 45(12): 3100-3104.

[8] He B, Wang G. Is ceramsite the last straw for sewage sludge disposal: a review of sewage sludge disposal by producing ceramsite in China[J]. Water Science and Technology, 2019, 80(1): 1-10.

[9] Rostami F, Tafazzoli S M, Aminian S T, et al. Comparative assessment of sewage sludge disposal alternatives in Mashhad: a life cycle perspective[J]. Environmental Science and Pollution Research, 2020, 27(1): 315-333.

[10] Venkatesa Prabhu S, Hamda A S, Jayakumar M, et al. Production strategies and potential repercussion of sewage sludge biochar as a futuristic paradigm toward environmental beneficiation: a comprehensive review[J]. Environmental Quality Management, 2024, 33(3): 1-19.

[11] Li H, Shi Y, Wang Y, et al. A research on the strengthening effect of sludge charcoal on activated sludge process in sewage treatment[J]. Environmental Science and Pollution Research, 2024, 31(4): 5289-5303.

[12] 陈治池, 何强, 蔡然, 等. 碳中和趋势下数学模拟在污水处理系统中的发展与综合应用[J]. 中国环境科学, 2022, 42(06): 2587-2602.

[13] 和笑天, 王宏武, 李鹏飞, 等. AAO工艺大型城市污水处理厂春节期间脱氮优化技术[J]. 净水技术, 2025, 44(02): 208-215.

[14] Perez Adassus M B, Heffner H, López-Corral I, et al. Studies on elimination of nutrients from aqueous effluents using ZnO nanoparticles: the case of ammonium as a model. Experimental and theoretical insights[J]. Journal of Materials Science, 2024, 59(8): 3363-3380.

[15] Behera C R, Srinivasan B, Chandran K, et al. Model based predictive control for energy efficient biological nitrification process with minimal nitrous oxide production[J]. Chemical Engineering Journal, 2015, 268: 300-310.

[16] LI Y, ZHOU L W, WANG R Z. Urban biomass and methods of estimating municipal biomass resources[J]. Renewable and Sustainable Energy Reviews, 2017, 80: 1017-1030.

[17] Schmitt F, Khac-Uan D. Prediction of membrane fouling using artificial neural networks for wastewater treated by membrane bioreactor technologies: bottlenecks and possibilities[J]. Environmental Science and Pollution Research, 2017, 24(29): 22885-22913.

[18] Tosh C R, Mcnally L. The relative efficiency of modular and non-modular networks of different size[J]. Proceedings of the Royal Society B-Biological Sciences, 2015, 282(1802): 20142568.

[19] Amer M, Maul T. A review of modularization techniques in artificial neural networks[J]. Artificial Intelligence Review, 2019, 52(1): 527-561.

[20] Xie Y, Wen Z, MO Z, et al. Implementation of an automatic and miniature on-line multi-parameter water quality monitoring system and experimental determination of chemical oxygen demand and ammonia-nitrogen[J]. Water Science and Technology, 2016, 73(3): 697-706.

[21] Li D, Xu X, Li Z, et al. Detection methods of ammonia in water: A review[J]. Trac-Trends in Analytical Chemistry, 2020, 127: 115890.

[22] Shi C, Xu Z, Smolinski B L, et al. Spectrophotometric analyses of hexahydro-1, 3, 5-trinitro-1, 3, 5-triazine (RDX) in water[J]. Journal of Environmental Sciences, 2015, 33: 39-44.

[23] Papias S, Masson M, Pelletant S, et al. In situ continuous monitoring of nitrogen with ion-selective electrodes in a constructed wetland receiving treated wastewater: an operating protocol to obtain reliable data[J]. Water Science and Technology, 2018, 77(6): 1706-1713.

[24] Mutlu E, KURNAZ A. Assessment of physicochemical parameters and heavy metal pollution in Çeltek Pond water[J]. 2018, 47(6):1185-1192.

[25] 和嘉伟, 杨泽明, 赵金成, 等. 河口岸带氨氮原位快速分析方法研究[J]. 热带海洋学报, 2024, 43(04): 144-152.

[26] Xing Z, Lin J, McCoul D, et al. Inductive strain sensor with high repeatability and ultra-low hysteresis based on mechanical spring[J]. IEEE Sensors Journal, 2020, 20(24): 14670-14675.

[27] Zhang C, Dou D, Peng Y, et al. Froth-tailings fusion image-based concentrate ash content prediction in coal flotation using stacking model[J]. Energy Sources Part a-Recovery Utilization and Environmental Effects, 2025, 47(1): 31-48.

[28] Schaedler M, Böcherer G, Pittalà F, et al. Recurrent neural network soft-demapping for nonlinear ISI in 800Gbit/s DWDM coherent optical transmissions[J]. Journal of Lightwave Technology, 2021, 39(16): 5278-5286.

[29] Liu S, Gao X, Qi W, et al. Soft sensor modelling of propylene conversion based on a Takagi-Sugeno fuzzy neural network optimized with independent component analysis and mutual information[J]. Transactions of the Institute of Measurement and Control, 2019, 41(3): 737-748.

[30] 杨琴, 谢淑云. BP神经网络在洞庭湖氨氮浓度预测中的应用[J]. 水资源与水工程学报, 2006, (01): 65-70.

[31] 杨琴, 田永红. BP神经网络和遗传算法相结合在洞庭湖氨氮浓度预测中的应用[J]. 长江大学学报(自然科学版)理工卷, 2010, 7(04): 110-112+121+8.

[32] Wang G, Jia Q-S, Zhou M C, et al. Soft-sensing of Wastewater Treatment Process via Deep Belief Network with Event-triggered Learning[J]. Neurocomputing, 2021, 436: 103-113.

[33] Bi J, Zhang L, Yuan H, et al. Multi-indicator water quality prediction with attention-assisted bidirectional LSTM and encoder-decoder[J]. Information Sciences, 2023, 625: 65-80.

[34] Wurzberger F, Schwenker F. Learning in deep radial basis function networks[J]. Entropy, 2024, 26(5): 368.

[35] Zhang Y, Zheng Q, Chen X, et al. Comparison and analysis of several quantitative identification models of pesticide residues based on quick detection paperboard[J]. Processes, 2023, 11(6): 1854.

[36] Li M, Li W, Qiao J. Design of a modular neural network based on an improved soft subspace clustering algorithm[J]. Expert Systems with Applications, 2022, 209: 118219.

[37] Li W, Li M, Qiao J, et al. A feature clustering-based adaptive modular neural network for nonlinear system modeling[J]. ISA transactions, 2020, 100: 185-197.

[38] Fu Z, Pan X. Robust and Accurate Recognition of Carriage Linear Array Images for Train Fault Detection[J]. Applied Sciences, 2024, 14(18): 8525.

[39] Hendel M, Bousmaha I S, Meghnefi F, et al. An Intelligent Power Transformers Diagnostic System Based on Hierarchical Radial Basis Functions Improved by Linde Buzo Gray and Single-Layer Perceptron Algorithms[J]. Energies, 2024, 17(13): 3171.

[40] Georgescu P L, Moldovanu S, Iticescu C, et al. Assessing and forecasting water quality in the Danube River by using neural network approaches[J]. Science of the Total Environment, 2023, 879: 162998.

[41] 乔俊飞, 安茹, 韩红桂. 基于RBF神经网络的出水氨氮预测研究[J].控制工程, 2016, 23(09): 1301-1305.

[42] 杨壮, 武利, 乔俊飞. 基于GM-RBF神经网络的污水环境预测[J]. 控制工程, 2019, 26(09): 1728-1732.

[43] 乔俊飞, 安茹, 韩红桂. 基于相对贡献指标的自组织RBF神经网络的设计[J]. 智能系统学报, 2018, 13(02): 159-167.

[44] Jacobs R A, Jordan M I, Nowlan S J, et al. Adaptive mixtures of local experts[J]. Neural computation, 1991, 3(1): 79-87.

[45] Huang W, Yang L, Zhan X, et al. Synchronization transition of a modular neural network containing subnetworks of different scales[J]. Frontiers of Information Technology & Electronic Engineering, 2023, 24(10): 1458-1470.

[46] Lu B L, Ito M. Task decomposition and module combination based on class relations: a modular neural network for pattern classification[J]. IEEE transactions on neural networks, 1999, 10(5): 1244-1256.

[47] Li K, Zhang Z, Yu Z. Modular stochastic configuration network with attention mechanism for soft measurement of water quality parameters in wastewater treatment processes[J]. Information Sciences, 2025, 689: 121476.

[48] Qiao J F, Meng X, Li W J, et al. A novel modular RBF neural network based on a brain-like partition method[J]. Neural Computing and Applications, 2020, 32(3): 899-911.

[49] Meng X, Zhang Y, Qiao J. An adaptive task-oriented RBF network for key water quality parameters prediction in wastewater treatment process[J]. Neural Computing and Applications, 2021, 33(17): 11401-11414.

[50] 张昭昭, 潘浩然, 朱应钦. 注意力改进的动态自组织模块化神经网络结构设计及应用[J]. 计算机科学, 2024, 51(S2): 173-181.

[51] 张昭昭, 乔俊飞, 余文. 多层自适应模块化神经网络结构设计[J]. 计算机学报, 2017, 40(12): 2827-2838.

[52] 郭鑫, 李文静, 乔俊飞. 基于自组织模块化神经网络的污水处理过程出水参数预测[J]. 化工学报,2024, 75(09): 3242-3254.

[53] Meng X, Tang J, Qiao J. NOx emissions prediction with a brain-inspired modular neural network in municipal solid waste incineration processes[J]. IEEE Transactions on Industrial Informatics, 2021, 18(7): 4622-4631.

[54] 贾丽杰, 李文静, 乔俊飞. 基于神经元特性的径向基函数神经网络自组织设计方法[J]. 控制理论与应用, 2020, 37(12): 2618-2626.

[55] 张伟, 黄卫民. 基于SAPSO算法的RBF神经网络设计[J]. 控制与决策, 2021, 36(09): 2305-2312.

[56] Jia L, Li W, Qiao J. An online adjusting RBF neural network for nonlinear system modeling[J]. Applied Intelligence, 2023, 53(1): 440-453.

[57] 关李晶, 何洁帆, 张立勇, 等. 基于单输出子网迭代学习的缺失值填补方法[J]. 大连理工大学学报, 2022, 62(04): 427-432.

[58] Wang Y, Sun S, Zhong J. An ensemble anomaly detection with imbalanced data based on robot vision[J]. Int. J. Rob. Autom, 2016, 31(2):77-83.

[59] Bera S P, Godhaniya M, Kothari C. Emerging and advanced membrane technology for wastewater treatment: A review[J]. Journal of Basic Microbiology, 2022, 62(3-4): 245-259.

[60] Vendramelli R A, Vijay S, Yuan Q. Mechanism of nitrogen removal in wastewater lagoon: a case study[J]. Environmental technology, 2017, 38(12): 1514-1523.

[61] Zhang Y, Pang B, Bo G, et al. Deep Denitrification of the Simulated Coastal Secondary Effluent with High Chloride Ion by Electrolysis[J]. Polish Journal of Environmental Studies, 2021, 30(3):2437-2443.

[62] Tong W, Liu S, Gao X Z. A density-peak-based clustering algorithm of automatically determining the number of clusters[J]. Neurocomputing, 2021, 458: 655-666.

[63] Enns J T, Coren S. The box alignment illusion: An orientation illusion induced by pictorial depth[J]. Perception & psychophysics, 1995, 57(8): 1163-1174.

[64] Qiao J, Guo X, Li W. An online self-organizing modular neural network for nonlinear system modeling[J]. Applied Soft Computing, 2020, 97: 106777.

[65] Qiao J, Zhang Z, Bo Y. An online self-adaptive modular neural network for time-varying systems[J]. Neurocomputing, 2014, 125: 7-16.

[66] Chandra R, Ong Y S, Goh C K. Co-evolutionary multi-task learning for dynamic time series prediction[J]. Applied Soft Computing, 2018, 70: 576-589.

[67] Han H G, Guo Y N, Qiao J F. Nonlinear system modeling using a self-organizing recurrent radial basis function neural network[J]. Applied soft computing, 2018, 71: 1105-1116.

[68] Juang C F, Lin Y Y, Tu C C. A recurrent self-evolving fuzzy neural network with local feedbacks and its application to dynamic system processing[J]. Fuzzy Sets and Systems, 2010, 161(19): 2552-2568.

[69] 郭鑫, 李文静, 乔俊飞. 一种改进的在线自适应模块化神经网络[J]. 控制与决策, 2020, 35(07): 1597-1605.

[70] 蒙西, 乔俊飞, 韩红桂. 基于类脑模块化神经网络的污水处理过程关键出水参数软测量[J]. 自动化学报, 2019, 45(05): 906-919.

[71] Yilmaz S, Oysal Y. Fuzzy wavelet neural network models for prediction and identification of dynamical systems[J]. IEEE transactions on neural networks, 2010, 21(10): 1599-1609.

[72] 郭鑫, 李文静, 乔俊飞. 基于自组织模块化神经网络的污水处理过程出水参数预测[J]. 化工学报, 2024, 75(09): 3242-3254.