原位污泥减量新工艺反应机制及其基于机器学习的智能控制研究

At present, the problems of urban sewage treatment and excess sludge disposal in China become very prominent. Thus, the development of simultaneous enhanced sewage treatment and in-situ excess sludge reduction process, and the realization of its stable operation in an actual sewage treatment plant, become the urgent and significant issues in the construction and reconstruction in the existing sewage treatment plants. In this work, an integrated A2MO-M+GAC/MBR system (abbreviate for A2MMBR-M system) was developed and the running parameters of simultaneous enhanced sewage treatment and in-situ sludge reduction effects were optimized. The mechanism of in-situ sludge reduction and the main influencing parameters were investigated, and the transformation and mechanisms of organic matters, nitrogen and phosphorus removal in the A2MMBR-M system were further clarified. From the point of view of microbial ecology, the molecular ecological mechanisms of enhanced sewage treatment and in-situ sludge reduction effects were studied. The interaction relationships between the EPS generation and degradation and the sewage treatment process (including organic treatment and biological phosphorus removal mechanisms) in the A2MMBR-M system were investigated, and an improved ASM2d-EPS mathematical model was developed. Further researches on the smart control strategies by combining ontology theory and ASM2d mathematical model to improve the Q-learning algorithm in the A2MMBR-M system were carried out. It is of great scientific value and creative significance to realize the smart running of the existing and the new developed sewage treatment systems based on Machine Learning theory.

目前，我国城市污水及污泥的处理处置问题已十分突出，开发并建立强化污水处理与原位污泥减量耦合工艺，并实现在实际工程中的稳定运行是我国污水厂新建及改造的重点和难点。本研究以自主开发的一体式A2MO-M+颗粒活性炭/膜生物工艺（A2MMBR-M）为研究对象，优化系统原位污泥减量及强化污水处理的调控对策；明晰A2MMBR-M系统污泥减量的主导因素及反应机理；阐明A2MMBR-M系统有机污染物、氮、磷的降解规律及作用机制，明晰系统的分子生态学机制；揭示A2MMBR-M系统中微生物胞外聚合物（EPS）的生成代谢与污水处理过程的互作关系，建立改进的ASM2d-EPS数学模型；并进一步开展结合本体论理论和ASM2d模型的Q-learning改进算法在A2MMBR-M系统的智能化调控策略的研究，对全面实现基于机器学习的新型污水处理系统及现有污水厂在实际工程中的智能化运行具有重要的科学价值和创新意义。

本研究结合大数据及机器学习算法，开展了活性污泥污水处理工艺及原位污泥减量过程优化运行及智能调控研究，主要研究内容如下：. 1）开发了一种污泥隐性生长预处理耦合的缺氧/缺氧工艺系统（厌氧-好氧+碱/超声，AO+ALK/ULS），并通过BP人工神经网络（BPANN）模型及Levenberg-Marquardt算法对该耦合工艺进行实时响应和优化控制，以提高低碳/氮（C/N）比的生活废水中生物脱氮（BNR）。研究结果表明BPANN模型有效地建立了AO+ALK/ULS系统中相关输入变量与出水输出之间复杂的非线性关系，有效地实现了实时动态响应和过程优化控制，建立了同步提高BNR和原位污泥还原方面的新的优化建模方法，该模型可实现现有污水处理工艺对低C/N生活废水的实时动态响应和工艺优化控制；. 2）搭建了基于Q学习（QL）的传统厌氧-好氧（AN/AE，AO）污水处理厂整体仿真模型，并结合实际真实污水厂进出水水质及其对应控制方式，提出了基于QL的污水处理过程智能优化算法，建立了QL算法在不同水力停留时间（HRT）步长下得到的Q矩阵和状态转移矩阵，利用Q矩阵和<当前状态-转移状态>序偶，实现了AO系统中的AN/AE过程中HRT的优化，从而建立了基于QL的HRT实时、动态优化控制策略框架，进而实现了任意强化生物除磷（EBPR）系统中AN/AE过程，任意进水水质条件下的最优HRT组合。研究结果表明所提出的基于QL的AO系统优化控制模型运行良好，所得到的QL算法成功地实现了污水处理过程实时优化控制建模，进而保证在进水负荷波动的情况下具有稳定的最优控制策略；. 3）提出了一种基于活性污泥No. 2d数学模型（ASM2d）改进的QL算法（ASM2d-QL），以厌氧-缺氧-好氧（AAO）工艺为例，通过最终出水水质最优的目标函数建立<回报-动作>序偶，分别得到控制策略（HRT和内循环率（IRR））对应的优化Q矩阵，进而实现优化后AAO工艺的控制策略具有自适应能力，从而推导了具有自学习机制的ASM2d-QL集成算法，可以实现AAO系统在不同进水负荷下的控制策略实时、智能响应。本研究可在进水负荷波动的情况下，很好地实现智能建模并输出稳定的最优控制策略。