污水处理过程多模态建模及相应故障预测方法研究

Soft-sensors are considered a good alternative to predict hard-to-measure variables in the wastewater process. One of the bottlenecks limiting its widespread application, however, is that modals (operating modes) change frequently. The objective of this project is to predict safety-related but hard-to-measure variables in the wastewater process. Firstly, in order to make sure the accuracy of soft-sensors prediction under normal and fault modals, Enhanced Rao-Blackwellised particle filter, Adaptive Neuro-Fuzzy Inference System (ANFIS) together with Probabilistic Hybrid Automata are coordinated to build a multi-modal soft-sensor model. Secondly, for the purpose of exploring the rule of bacteria inactivation in sludge and constructing a multi-step ahead prediction model under the fault modal, a high-order Markov model with ANFIS structure is performed. Finally, on this basis, improved Roll Time Series and sensitivity analysis from statistical methods are further implemented accounting for the prognostic uncertainties and improving the prediction accuracy. This project will explore a multi-modal soft-sensor modeling methodology and its corresponding prognostic functions, these methods will be better guidance for monitoring system operation, it will be also make a greater contribution for system health management in a quantitative manner theoretically and technically.

软测量技术作为解决污水处理过程难以实时测量变量的有效途径正受到广泛的关注。然而，模态的频繁变化成为了制约该项技术推广应用的一个主要瓶颈。项目拟以污水处理中与安全密切相关但又难以测量的过程变量为研究对象，首先依据改进Rao-Blackwellised粒子滤波理论，自适应神经模糊推理模型理论和概率混合自动机理论，开展多模态软测量建模方法的研究，保证软测量在正常模态和故障模态等多种模态都能做出准确预测；其次，在污泥中毒故障模态下，研究以自适应神经模糊推理为模型结构的高阶马尔科夫多步预测，探索污泥中毒退化规律和建立故障预测模型；最后，在上述研究的基础上，结合改进滚动式时间序列模型和统计学中敏感性分析方法，对故障预测不确定性进行深入分析从而提高预测的准确度。该项目将探索多模态软测量建模方法及相应的故障模态下故障预测功能，旨在为有效的监控系统运行和定量评价系统健康状态奠定理论和技术基础。

为了解决模态的频繁变化制约软测量技术推广应用，项目以污水处理中与安全密切相关但又难以测量的过程变量为研究对象，首先改进了ARMA模型，高斯过程模型和自适应神经模糊推理模型理论，并与概率混合自动机理论相结合，实现了软测量在正常模态和故障模态等多种模态都能做出准确预测；其次，在污泥中毒故障模态下，研究以ARMA模型，高斯过程模型和自适应神经模糊推理模型的多步预测，探索污泥中毒退化规律和建立故障预测模型；最后，在上述研究的基础上，结合改进滚动式时间序列模型和统计学中敏感性分析方法，对故障预测不确定性进行深入分析从而提高预测的准确度。项目探索了污水系统运行规律，合理地描述污水处理系统复杂多变的多模态特性；对故障模态下不确定性进行充分认识和量化的基础上实现准确的故障预测。此外，利用MATLAB软件开发了一款针对污水处理多模态软测量建模和故障预测的软件，可实现多模态建模和故障预测等功能，并在国际水协会的仿真平台以及实际采集的数据上进行了测试。通过本项目的研究，建立了一套多模态建模和故障预测的方法和理论体系，丰富相关的理论和内容。