污水处理厂涉及复杂的生物化学反应， 具有强非线性、耦合性、不确定性和时变性特点，难以对未来水质信息进行准确预测以辅助污水处理厂实现前馈控制； 此外，污水处理厂管理者倾向于增加设备运行负荷以防止污染，从而导致运行能耗较高。 针对污水处理厂难以预测水质导致工艺调整滞后以及运行能耗和水质难以权衡问题，提出了城市污水处理厂水质预测及工艺参数多目标优化研究。 主要研究工作如下：

（1）针对污水处理厂水质波动较大难以实现准确预测的难点， 开展了基于分解集成框架的水质预测研究，实现水质有效点预测及区间预测。 具体来说， 首先基于改进变分模态分解将原始水质数据分解为包含更多信息和更少噪声的子序列；其次根据子模型选择策略，选出每个子系列的最优子模型， 并结合子模型的预测结果得到最终点预测结果；最后基于自适应核密度估计构建稳健可靠的预测区间。 本文以武汉某水厂水质参数对研究进行验证，对于出水总氮的预测精度 R2 可达 0.955；基于改进变分模态分解相较于其它分解策略有着最好预测性能；基于自适应核密度估计构建的区间在不同置信水平和优化目标下 CWC 值最小，表明其构建的区间准确且可靠。

（2）针对污水处理厂能耗与水质难以权衡以及工艺调整滞后的问题， 开展了基于时序预测的污水处理工艺参数多目标优化研究，求解未来入水所对应的工艺参数的设定值以实现节能和保质目标。 具体来说， 首先基于极端梯度提升以构建优化目标的预测模型；其次基于分解集成策略对入水水质的预测； 最后引入多目标进化算法及伪权重对问题进行优化决策。 本文基于基准仿真平台对研究进行验证，基于优化的极端梯度提升可以使优化目标的预测精度 R2 达 0.90 以上；基于分解集成框架对入水水质预测精度 R2 达 0.95 以上；最后基于多目标进化算法和伪权重法对问题进行优化求解，在能耗与水质权重相等时，能耗优化效率可达 2.96%，且此时水质优化率达 10.53%。
本文开展的城市污水处理厂水质预测及工艺参数多目标优化研究， 能准确可靠地预测水质， 并求解未来入水情况对应的工艺参数优化值， 实现对工艺流程前馈控制， 以实时降低运行能耗且保证水质达标，有助于污水处理厂实现可持续发展目标。

关键词：污水处理厂；时间序列预测；多目标优化；时间序列分解；污水处理建模

Wastewater treatment plants (WWTPs) involve complex biochemical reactions characterized by strong nonlinearity, coupling, uncertainty, and time-varying characteristics, which make it difficult to accurately predict future water quality information to assist WWTPs in realizing feed-forward control; additionally, WWTPs’ managers tend to increase the operating load of the equipment to prevent pollution, which leads to higher operational energy consumption. Aiming at the problems of lagging process adjustment due to the difficulty of water quality prediction in WWTPs, and the difficulty of trade-off between operational energy consumption and water quality, this paper proposes a study on water quality prediction and multi-objective optimization of process parameters in WWTPs. The main research work is as follows:
(1) Aiming at the difficulty of accurate prediction of water quality fluctuation in WWTPs, the study of water quality prediction based on the decomposition integration framework is carried out to realize the effective point prediction and interval prediction. Specifically, firstly, the original water quality data are decomposed into sub-series containing more information and less noise based on the improved variational modal decomposition; secondly, the optimal sub-model for each sub-series is selected according to the sub-model selection strategy and the final point prediction result is obtained by combining the prediction results of the sub-models; and finally, the robust and reliable prediction intervals based on the adaptive kernel density estimation are constructed. This paper validates the study with the water quality parameters of a water plant in Wuhan. The prediction accuracy (R2) of total nitrogen in the effluent water is up to 0.955; based on the improved variational modal decomposition, it has the best prediction performance compared with other decomposition strategies; the interval constructed based on the adaptive kernel density estimation has the smallest value of CWC under different confidence levels and optimization objectives, which shows that the constructed intervals are accurate and reliable.
(2) Aiming at the problems of difficult trade-off between energy consumption and water quality in WWTPs and lagging process adjustment, the study on multi-objective optimization of wastewater treatment process parameters based on time series prediction is carried out to solve the set values of process parameters corresponding to future influent water to achieve the goals of energy-saving and quality preservation. Specifically, firstly, the prediction models of optimization objectives are constructed based on extreme gradient enhancement; secondly, the prediction of influent water quality is based on the decomposition and integration strategy; finally, the multi-objective evolutionary algorithm and pseudo-weights are introduced to optimize the problem and make decisions. This paper is based on the benchmark simulation platform to verify the research. Based on the extreme gradient enhancement of optimization, the prediction accuracy of the optimization objectives can reach more than 0.90; based on the decomposition and integration framework, the prediction accuracy (R2) of influent water quality can reach more than 0.95; finally, based on the multi-objective evolutionary algorithm and pseudo-weighting method to optimize and solve the problem, and when the weights of the energy consumption and the water quality are equal, the efficiency of the energy consumption optimization can be up to 2.96%, and the rate of the optimization of the water quality can be up to 10.53%.
The water quality prediction and multi-objective optimization of process parameters of urban WWTPs carried out in this paper can accurately and reliably predict the water quality and solve the optimized values of process parameters corresponding to the future influent situation, to realize the feed-forward control of the process to reduce the operational energy consumption in real time and ensure that the water quality meets the standard, which can help the WWTPs to achieve the goal of sustainable development.