带有工况特征识别的活性污泥过程水质参数稳定软测量方法研究

Water quality is the most important index for the environmental monitoring department， and accurate predictive model for water quality guarantees the stable operation and optimal control of wastewater treatment plant(WWTP). This project focuses on the accuracy and adaptivity of water quality soft sensor under uncertain disturbances and fluctuant operational conditions. Based on online and offline data combining expert knowledge,extraction methods of qualitative characteristics for sludge bulking based on qualitative trend analysis are used to recognize abnormal working conditions; adaptive hybrid soft sensor of water qualtiy is composed of simplified mechanism model of activated sludge process used as main model and RBF neural network with stable learning algorithm as error compensation model. The main contents of this project are as follows:(1)recognition of abnormal working conditions;(2)stable adaptive learning algorithm of hybrid soft sensor under normal working conditions;(3)stability and predictive performance analysis of soft sensor. Recognition of abnormal working conditions can improve the predictive performance under abnormal wording condition. Adaptive stable learning algorithm eliminates the influence of unmodeled dynamics and uncertain disturbances to model accuracy, and improves the model robustness. The achievement of this project produces profound and.lasting significance for the application of water quality soft sensor on advanced control system of wastewater treatment process.

污水水质是环境监测部门最关注的指标信息，准确的水质预测模型是污水处理厂稳定运行和优化控制的重要条件。本项目对活性污泥过程在未知外部干扰和运行条件波动时水质模型精度和自适应性展开研究，在污泥膨胀在线和离线数据的基础上，结合专家知识，采用基于定性趋势分析的污泥膨胀定性特征提取和识别方法，识别异常工况；以简化机理模型作为主模型，带稳定学习算法的神经网络模型作为误差补偿模型，建立正常工况下的水质自适应混合模型。主要内容包括：(1)异常工况识别方法；(2)正常工况下水质混合模型自适应稳定学习算法及稳定性分析；(3)水质模型稳定性及预测性能分析。异常工况识别可提高异常情况下的水质预测性能；稳定学习算法能降低未建模动态和不确定干扰对水质模型精度的影响，提高模型鲁棒性；误差补偿模型采用流形学习算法进行输入变量的有效特征提取。本项目对推动水质模型在污水处理过程高级控制系统中的应用具有深远的实际应用价值。

由于衡量污水处理过程出水质量的关键水质指标在线不可测或测量滞后非常大，在线测量仪器操作复杂，严重限制了简单、有效控制器的应用，水质闭环控制难以实现，尤其是当进水特性、运行条件等生产边界条件发生变化时，仅仅依靠人的经验及传统技术难以满足愈来愈高的优化目标要求。为了寻求污水处理过程实时优化控制策略，改善污水处理过程的出水水质、加强运行监控，进而降低生产能耗，需要解决污水关键水质指标的在线检测和控制问题，重点在于如何提高水质软测量的精度与控制器的鲁棒性。.项目主要研究内容：① 在污水处理过程异常工况识别方法；② 在污水处理过程正常工况下污水水质COD软测量方法；③ 针对污水处理过程的不确定性、未知扰动的特点，设计了鲁棒异步切换模型预测控制器。.在SVI模糊神经网络模型建模方面，所提出的自组织递归模糊神经网络在规则化层根据网络性能进行结构调整，与其他自组织算法比较表明，该算法测试误差至少可减少13.3%。在污水水质集成软测量模型建模方面，提出的基于简化机理模型和自组织RBF神经网络的集成模型具有较好的自适应性和鲁棒性。在鲁棒预测控制器设计方面，控制器的误差平均值为1.2823，而线性控制器方法的误差平均值为2.6466，控制性能大幅度提升。.本项目研究成果对于加速我国城市污水处理厂的水质监测和控制有良好的应用前景。

内容获取失败，请点击重试