摘要
自来水厂的混凝沉淀过程受水力条件、原水水质、絮凝剂投加量、水温、pH等多个因素影响,具有非线性、大时滞、多扰动等特点。传统的人工投加方式依赖主观经验,难以根据环境和水质变化动态准确地调整絮凝剂投加量,容易引发沉淀池出水水质不稳定或药物过量投加等问题。为此,研发了一种由分布式水下监测平台与时序神经网络预测模型结合PID控制算法构成的絮凝剂精准投加控制系统。首先,基于机器视觉的图像分割算法分析矾花颗粒面积占比和颗粒形态学特征,通过多维度综合研判絮凝反应效果,并对絮凝效果偏差的情况提供预警功能;其次,基于数据库中长期的监测数据,迭代训练时序神经网络预测模型,对出水浊度进行多步超前预测,突破了絮凝沉淀过程的大时滞响应与多因素扰动对自动控制算法带来的挑战。结果表明,采用精准投加控制系统后,出水浊度的波动幅度降低了62%,絮凝剂投加量减少了25%,在保证出水水质稳定的前提下,可以实现大幅降低药剂投加量的目标。
The coagulation and precipitation process in waterworks is affected by many factors such as hydraulic conditions,raw water quality,flocculant dosage,water temperature and pH,and has the characteristics of nonlinear,large time delay and multi-disturbance.The traditional manual dosing method relies on subjective experience,so it is difficult to adjust the dosage of flocculant dynamically and accurately according to the changes in environment and water quality,which is easy to cause problems such as unstable water quality of sedimentation tank or excessive dosing of chemicals.Therefore,a flocculant precise dosing control system composed of distributed underwater monitoring platform and time-series neural network combined with PID control algorithm was developed.Firstly,the area ratio and morphological characteristics of the flocculent particles were analyzed by the image segmentation algorithm based on machine vision,the flocculation reaction performance was comprehensively evaluated through the multidimensional characteristics,and an early warning function for the deviation of flocculation was provided.Secondly,based on the medium and long term monitoring data in the database,the time-series neural network prediction model was iteratively trained to conduct multi-step advance prediction of effluent turbidity.The challenge of the automatic control algorithm brought by the large time delay response and multi-factor disturbance in the flocculation and precipitation process was overcame.After adopting the precise dosing control system,the fluctuation range of effluent turbidity was reduced by 62%,and the dosage of flocculant was reduced by 25%.On the premise of ensuring the stability of effluent quality,the system achieved the goal of significantly reducing the dosage of chemicals.
作者
苏宇宸
李鹏飞
张新运
徐晓兵
季梦奇
SU Yu-chen;LI Peng-fei;ZHANG Xin-yun;XU Xiao-bing;JI Meng-qi(College of Mechanics and Materials,Hohai University,Nanjing,210098,China;Dongtai Water Supply Co.Ltd.,Dongtai,224200,China;Institute of Artificial Intelligence,Beihang University,Beijing 100191,China)
出处
《中国给水排水》
CAS
CSCD
北大核心
2023年第3期56-61,共6页
China Water & Wastewater
关键词
时序神经网络
絮凝监测
精准投加
机器视觉
出水浊度
time-series neural network
flocculation monitoring
precise dosing
machine vision
effluent turbidity
