在满足污水处理厂出水标准的条件下,为降低曝气能耗,研究提出高斯过程回归(GPR)联合改进黑寡妇算法(IBWO)的曝气优化策略。利用GPR建立进出水水质与曝气流量的关系模型,预测出水水质;基于关系模型构造出水水质概率约束的曝气能耗最小化问题,采用IBWO求解最优运行策略,并验证该策略运用于污水处理系统的改善效果。试验结果表明:在外部负荷波动的情况下,出水水质能够维持在设定限制值以下;精确的曝气控制显著降低了曝气能耗,相较于传统的粗放式操作,能源效率得到有效提升。

[Objective] Aeration represents the most energy-intensive process in wastewater treatment plants (WWTPs), accounting for over 50% of total energy consumption. Traditional operation strategies often rely on excessive aeration to ensure effluent compliance, causing substantial energy waste and increased carbon emissions. To address this challenge, we propose a data-driven probabilistic optimization framework integrating Gaussian process regression (GPR) with an improved black widow optimization (IBWO) algorithm to minimize aeration energy consumption while ensuring effluent quality compliance under dynamic and uncertain influent conditions. The novelty lies in combining a probabilistic prediction model with an enhanced evolutionary algorithm for adaptive and energy-efficient aeration control at the plant scale.[Methods] A GPR model was developed to describe the nonlinear relationships between influent characteristics, aeration flow rate, and effluent water quality indicators, and furthermore, an IBWO algorithm was constructed by enhancing the standard black widow optimization framework. The GPR-IBWO framework was executed in a closed-loop configuration, where the GPR model continuously updated effluent predictions from real-time process data, and IBWO dynamically optimized aeration flow rates to minimize energy consumption while maintaining effluent quality within probabilistic bounds. Benchmark tests on six CEC2017 functions demonstrated that IBWO outperformed GA, PSO, ABC, and standard BWO in convergence speed and solution accuracy. The GPR-IBWO strategy was tested in a full-scale municipal WWTP in Jiujiang City, China. Operational data were collected over 30 days during wet and dry seasons to capture hydraulic and load fluctuations. The performance of four strategies was compared: the proposed GPR-IBWO optimization, conventional plant control mode, baseline GPR-BWO strategy method, and benchmark LSTM-PSO framework. [Results] Under both hydrological conditions, GPR-IBWO consistently maintained effluent concentrations of N ${\mathrm{H}}_{4}^{+}$-N, TN, TP, COD, BOD₅, and SS below discharge limits, despite large influent quality variations. Other optimization schemes occasionally caused transient exceedances, particularly in TN and TP during high-load periods. GPR-IBWO exhibited smoother fluctuations and faster recovery from disturbances, indicating stronger robustness and adaptability. These advantages stemmed from the probabilistic nature of GPR, which incorporated uncertainty into decision-making, and from the enhanced global search capacity of IBWO, which avoided premature convergence and ensured reliable optimization results. Energy analysis demonstrated the method’s superiority. Compared to the conventional control strategy, total aeration energy consumption decreased by 41.62% during the wet season and 29.86% during the dry season, outperforming GPR-BWO (27.7% and 24.8%) and LSTM-PSO (31.8% and 19.9%). The optimized aeration profiles dynamically adapted to fluctuating influent loads, minimizing energy input while maintaining effluent compliance. The GPR model achieved high prediction accuracy, with mean absolute percentage errors below 2% for most indicators and coverage rates exceeding 93% within the 95% confidence interval, confirming the reliability of probabilistic predictions for real-time optimization in stochastic environments. [Conclusion] This study develops and validates a probabilistic intelligent optimization framework for aeration control that integrates machine learning-based prediction with evolutionary optimization. The main conclusions are as follows. (1) The GPR-based model accurately captures nonlinear and uncertain process dynamics, providing high-confidence effluent predictions with narrow uncertainty intervals. (2) The IBWO algorithm, through multi-mechanism improvement, demonstrates superior convergence, robustness, and solution quality compared to mainstream metaheuristics. (3) The integrated GPR-IBWO strategy effectively maintains effluent quality compliance and reduces aeration energy consumption by up to 40%, confirming its strong adaptability to seasonal and hydraulic variations. (4) For long-term implementation, periodic recalibration of the GPR model or incorporation of disturbance observers is recommended to mitigate model drift and sustain optimization performance. Overall, the GPR-IBWO framework offers a generalizable and uncertainty-aware approach for optimizing WWTP operations, enhancing energy efficiency without compromising effluent quality and showing substantial potential for large-scale application in smart wastewater management and low-carbon environmental infrastructure.