基于机器视觉的水处理絮凝过程中絮体检测与控制模型研究

As for the characteristics of complex flocculation process, different influential factors, nonlinearity and large time-delay, while the current study such as representative of sampling points of floc measurement being weak, tracking algorithm being complex, along with the difficulty of setting up control model in water treatment, the theory and methods of machine vision and sampling both in the flocculation basin and the sedimentation basin will be used to measure size and quantity of floc. Floc movement track equation will be established in the sedimentation basin based on the Fluent and floc sedimentation velocity will be measured based on particle filter multi-target tracking algorithm. The flocculation mechanism will be further discovered based on the measurement results. The fusion model of multi-source data, which reveals the relations of dosage, raw water quality, floc size and quantity in the flocculation basin, floc sedimentation velocity in the sedimentation basin and the quality of precipitation water, will be set up based on dedifferential evolution(DE). Taking sedimentation velocity in the sedimentation basin as master control, and floc size, quantity in the flocculation basin as auxiliary control, a closed loop bunch-rank coagulant dosage control system, which both consider raw water quality change of real-time and precipitation water turbidity lagging, will be constructed. The research achievements will explore new ways for floc measurement and setting up control model of complex water treatment process, and have academic value in promoting flocculation theory and technology progress. The application of the research will have great realistic significance in improving water quality, ensuring the security of drinking water and reducing the run cost of waterworks.

针对水处理絮凝过程复杂，影响因素多，具有非线性、大滞后的特点，絮体检测采样点代表性不强，絮体跟踪算法复杂，控制模型难以建立等研究现状。采用机器视觉方法，絮凝池和沉淀池双池采样，对絮凝池检测絮体粒径和数量，基于计算流体力学Fluent软件建立沉淀池中絮体运动轨迹方程，采用粒子滤波多目标跟踪算法检测絮体沉速；基于絮体检测结果，进一步揭示絮凝机理；采用差分进化算法建立投药量与原水水质、絮凝池絮体粒径与数量、沉淀池絮体沉速、滤前水水质等参数关系的多源数据融合模型；以沉淀池中检测的沉降速度为主控，絮凝池中检测获取的粒径、数量等参数为辅控，构建一个既考虑原水水质变化的实时性又考虑滤前水浊度的滞后性的混凝投药闭环串级控制系统。研究成果可为复杂水处理过程絮体检测、控制模型的建立探索新途径，推动絮凝理论的发展，具有一定的学术价值。成果的应用对提高供水水质、保障饮用水安全、降低水厂运行成本具有重大的现实意义。

针对水处理絮凝过程复杂，影响因素多，具有非线性、大滞后的特点，絮体检测采样点代表性不强，絮体跟踪算法复杂，控制模型难以建立等研究现状，本项目研究主要内容及成果如下。.（1）基于机器视觉系统，提出了一种新的压缩感知与粒子滤波结合算法，获取了检测絮凝池絮体等效粒径、数量及沉速的有效识别及跟踪算法，实现了絮体等效粒径、数量和沉降速度的检测，能连续跟踪絮体并有效记录絮体航迹。.（2）编写了一种絮体图像数据处理软件Floc Processor-Microsoft Visual Studio，将机器视觉得到的各项絮体性能参数导入后分析并计算，检测在不同絮凝时间、混凝剂投药量的情况下对絮体性能参数及处理效果的影响，进一步揭示了絮凝机理。研究了絮体的数量、等效粒径和分形维数等性能参数与沉后水水质（包括浊度和ζ电位）之间的关系，如：沉后水浊度与絮体数量的关系式为y=0.00444x^2-0.95076x+51.42522，相关系数R^2为0.982。.建立了絮体数量、等效粒径、分形维数与沉后水浊度之间的多源数据融合模型，如：絮体的分形维数与等效粒径之间的相关关系式为Df=1.0224d^0.13147，符合幂指数关系，其相关系数R^2为0.826。.（3）基于遗传算法和BP神经网络前馈与PID反馈控制，采用了MATLAB仿真软件自主编码，构建了既考虑原水水质变化的实时性又考虑滤前水浊度滞后性的水处理混凝投药闭环串级控制系统。.研究成果为复杂水处理过程絮体检测、控制模型的建立探索了一条新途径，推动了絮凝理论的发展，具有一定的学术价值。