臭氧正日益广泛地被应用于城市饮用水处理，以强化水中天然有机物（NOM）的降解去除。对于膜法水厂，适量投加臭氧可以在提高NOM去除效果的同时，缓解膜污染。为了控制处理成本和能耗，有必要探究臭氧投加量的优化方案。本研究在文献调研的基础上，以不同浓度的腐殖酸模拟配水和天然水为处理对象，通过臭氧氧化批次实验和臭氧预氧化-微滤（MF）连续流实验，确定SUVA₂₅₄、膜污染指数（FI）和臭氧利用率（OUR）作为优化评价指标，进而探究了臭氧投加量对三个指标以及相关光谱参数的影响规律和机理。在此基础上，采用层次分析法，建立了一种基于多目标优化模型的臭氧投加量计算方法，并用于连续处理系统的在线前馈控制。实验结果表明，根据进水紫外光谱变化情况的实时分析数据，该计算方法可较为准确地预测膜法水处理的预臭氧投加量变化，优化膜过滤效果。本研究的主要发现和结论如下：

（1）批次实验发现，臭氧氧化能够破坏NOM中的紫外发色团，并将大分子组分分解为小分子组分，UV₂₅₄和光谱斜率的变化与上述两种变化分别有良好的相关性。紫外及荧光光谱结果显示，臭氧氧化降解了水中部分腐植酸和非腐殖质类天然有机物，并破坏了芳香型或共轭双键型紫外发色团，但对有机物的矿化度较低。排阻色谱分析结果表明，臭氧将大分子和中分子量有机组分分解为小分子量有机组分。常温和低温条件的批次实验结果显示，直接臭氧氧化速率常数kD随水中有机物浓度的增大和温度的降低而减小。SUVA₂₅₄能够反映NOM的反应性、亲疏水性和芳香性物质比例，可以作为臭氧氧化过程中NOM氧化效果的评价指标。分别对SUVA₂₅₄和OUR变化曲线的快、慢两个反应阶段进行线性拟合，两组直线交点对应的临界臭氧投加量具有良好相关性，因此，由SUVA₂₅₄确定的临界臭氧投加量可作为批次反应的最佳投加量。此外，比较各波段的光谱斜率，S_275~295与SUVA₂₅₄的拟合度最高，可以作为SUVA₂₅₄的替代参数，模拟配水（R² = 0.9427）得出的规律同样适用于天然水（R² = 0.9246）。

（2）连续流实验发现，臭氧预氧化显著缓解了微滤膜的可逆和不可逆膜污染，同时小幅降低了膜对DOC的截留效果，而S_275~295和UV₂₅₄可以分别作为SUVA₂₅₄和FI的替代评价参数。随着臭氧流量的增加，臭氧与膜过滤组合工艺的UV₂₅₄去除率逐渐上升，而总DOC去除率逐渐下降。通过排阻色谱分析，臭氧将大分子和中分子量有机组分分解为更易穿透膜孔的小分子量有机组分，因而降低了膜过滤截留效果。同时，臭氧预氧化将胶体有机物转化为溶解性有机物，降低了配水和天然水中NOM的膜污染趋势，使得可逆和不可逆膜污染均呈下降趋势，而可逆膜污染的缓解效果更明显。与批次实验结果相似，各波长范围的光谱斜率中，S_275~295与SUVA₂₅₄的拟合度最高（模拟配水R² = 0.9797，天然水R² = 0.9285）。同时，FI也与UV₂₅₄存在较高拟合度（R² = 0.8799），因此光谱参数（S_275-295和UV₂₅₄）可以作为组合工艺臭氧氧化效果指标的替代参数，用以简化水质检测和膜污染趋势评价。

（3）基于层次分析的多目标优化模型可用于臭氧最佳投加量计算，以实现臭氧预处理的在线前馈控制，同时水厂还可以在0.15~0.2 L·min^-1的臭氧流量下进行实验室小型批次实验，以确定最佳臭氧曝气流量。将SUVA₂₅₄、FI和OUR作为臭氧氧化效果指标，对其与臭氧投加量的关系曲线以负指数函数拟合，根据不同水厂的实际情况和需求，使用层次分析法建立对不同指标的判断矩阵和权重分配，以效果值量化臭氧氧化的整体效果。效果值峰值对应最佳臭氧投加量，进而得出最佳臭氧曝气流量。在此基础上，建立了不同的进水水质对应的最佳臭氧曝气流量的数据库，根据实时检测进水的UV₂₅₄改变臭氧曝气流量，实现臭氧投加的在线前馈控制。该前馈控制的精度指标CTI为0.01622，具有较高的准确性。此外，通过对比批次实验和连续实验所得的最佳臭氧流量，在0.15~0.2 L·min^-1的臭氧流量下，二者比值最接近1。同时模拟0.15L·min^-1臭氧流量下反应器内的流场，所得速度云图显示，两种进水方式下反应器内液体流速和气液混合情况相似。因此，无膜过滤单元的水厂可以在0.15~0.2 L·min^-1的臭氧流量下进行实验室小型批次实验来确定最佳臭氧投加量，进一步简化实验过程。

综上，本研究揭示了模拟配水和天然水中天然有机物的臭氧氧化机理，探明了光谱参数与臭氧氧化评价指标之间的关系，建立了臭氧投加的多目标优化模型和批次实验优化方法。所获成果为基于光谱检测的臭氧投加控制技术开发，及其在饮用水厂的应用提供了数据支持和技术指导。

Ozonation is increasingly used in municipal drinking water treatment plants to enhance the degradation and removal of natural organic matter (NOM) in the source water. For membrane-based treatment plants, adding ozone in an appropriate amount can improve NOM removal and mitigate membrane fouling at the same time. In order to control the cost and energy consumption of the treatment, it is necessary to develop suitable methodologies for ozone dosage optimization. In order to accomplish this goal, synthetic model waters containing different concentrations of humic acid (HA) and a natural surface water were treated with ozone oxidation, both in batch experiment and continuous preozonation-microfiltration (MF) experiment modes. It was found that SUVA₂₅₄, membrane fouling index (FI) and the ozone utilization rate (OUR) were suitable indicators for evaluating the efficacy of ozone oxidation, as a function of ozone dosage. Moreover, strong correlations were identified between the spectral parameters of the treated water and ozonation efficacy. Based on the aforementioned experimental results, a multi-objective optimization model with the analytic hierarchy process (AHP) was applied to determine the optimal ozone dose. The validity of this method was further assessed experimentally with online feedforward control of ozone dosing. The results showed that, through real-time analysis of UV spectrum of the source water, the changes in ozone dosage can be predicted and used to maintain the performance of the preozonation-MF system. Major findings and conclusions of this study are as follows:

(1) The results obtained in the batch experiments revealed that ozone oxidation mainly destroyed UV-related chromophores in NOM, and decomposed larger fractions into smaller ones. Also, the two effects were well correlated with the changes in UV₂₅₄ and the spectral slope of the treated water. As found by UV and fluorescence analyses, ozone oxidation partially degraded humic and non-humic NOM, destructed their aromatic or conjugated double bonds in chromophores, but the mineralization ratio was low. Size exclusion chromatography analysis showed that ozonation transformed larger organic fractions into smaller ones. The oxidation experiments conducted at normal and low temperature showed that the direct ozonation rate constant k_D decreased with increasing organic concentration in the water and temperature. Moreover, SUVA₂₅₄ reflects the reactivity, hydrophobicity and aromaticity of NOM, and therefore, can be used as an indicator to assess the oxidation efficacy of ozonation. Two-stage linear fitting of SUVA₂₅₄ and OUR vs. ozone dose curves yielded two pairs of fitting lines. The ozone doses corresponding to the two intersection points of the lines were similar, which suggested that the optimal dose determined by SUVA₂₅₄ can be used in the batch ozonation process. Besides, S_275-295 had the best goodness of fitting with SUVA₂₅₄, among the studied spectral slope factors. Thus, it can be adopted as a substitute parameter for SUVA₂₅₄, both for the synthetic model waters (R² = 0.9427) and the natural surface water (R² = 0.9246).

(2) Based on the continuous ozonation-MF experiments, preozonation significantly mitigated the reversible and irreversible membrane fouling, while slightly reduced dissolved organic carbon (DOC) retention by membrane filtration. S_275~295 and UV₂₅₄ can be used as substitute parameters for SUVA₂₅₄ and FI, respectively. With the increase in ozone flowrate, the combined ozonation-MF process gradually showed increases in the total percentage removal for UV₂₅₄, but decreases for DOC. Based on the HPSEC analysis, ozonation decomposed large organic fractions into small fractions that were easier to penetrate the membrane pores, thus significantly reducing DOC retention. At the same time, ozonation transformed colloidal organic matter into DOC, which reduced the fouling tendency of NOM. As a result, both reversible and irreversible membrane fouling were mitigated after ozonation, which the decrease in reversible fouling was more obvious. Similar to the findings obtained in the batch experiments, S_275~295 showed the best goodness of fitting with SUVA₂₅₄ (R² = 0.9797 for the synthetic model waters, and 0.9285 for the natural surface water). Also, a good correlation was found between FI and UV₂₅₄. Consequently, the two spectral parameters, S_275~295 and UV₂₅₄, may be used as simplistic substitute parameters for ozone oxidation assessment in continuous-flow conditions.

(3) The multi-objective optimization model can be applied for the prediction of optimal ozone dosing, for online feedforward dosing control. Alternatively, water treatment plants may conduct laboratory batch experiments at the ozone flowrate range of 0.15~0.2 L·min^-1, to determine the optimal ozone dose. Taking SUVA₂₅₄, FI and OUR as the indicators for ozone oxidation efficacy, their correlation curves with ozone dosage were fitted into negative exponential functions. According to the actual needs of different water treatment plants, AHP was further used to determine the judgment matrix and indicator’s weight, which were sequentially employed to calculate an effect value that quantified the overall effects of ozone oxidation. As such, the optimal ozone dose was determined based on the maximal effect value. In this way, an ozone dosing database was established for different source water conditions and treatment goals, and applied in feed-forward control of preozonation. This control method was shown to be accurate, with a CTI value of 0.01622. Moreover, a comparison of the optimal ozone flow rate obtained from the batch experiments and the continuous-flow experiment approached almost unity when the ozone flow rate was set at 0.15~0.2 L·min^-1. Corresponding simulation of the flow fields in the ozonation reactor under batch and continuous-flow conditions also revealed similar patterns for ozone and water mixing. Therefore, water treatment plants without the need for membrane fouling control may consider to carry out laboratory batch experiments at an ozone flowrate of 0.15~0.2 L·min^-1 to determine the optimal ozone dosage for full-scale application. This would further simplify the experimental process for determining the optimal ozone dose.

Overall, possible mechanisms for ozone oxidation of NOM present in the synthetic model waters and the natural surface water were revealed in this study, together with their resulting changes to the spectral characteristics of ozonated water. Two methods for ozone dose optimization were also established, on the bases of a multi-objective optimization model and the batch experiment, respectively. This study provided important baseline data and support for the development of spectroscopic technologies for ozone dosing control, as well their applications in drinking water treatment plants.