Research On Coagulant Dosing Control In Waterworks Based On Machine Learning

In the process of water treatment,the coagulation sedimentation process can effectively remove the suspended impurities and harmful substances in the water,and the timely adjustment of coagulant dosage can ensure the stability of the effluent quality of the coagulation sedimentation process.However,due to the influence of raw water quality indicators(such as raw water turbidity,water temperature,p H value,dissolved oxygen,oxygen consumption,etc.)and influent flow changes,especially the periodic,seasonal and time-varying changes of raw water quality,it brings great difficulties to coagulant dosing control.It can effectively ensure the water treatment effect of coagulation and sedimentation process and stabilize the effluent quality index(mainly effluent turbidity)of coagulation and sedimentation process by fully mining the big data information of raw water quality parameter changes and providing real-time reference for coagulant dosing control.Machine learning can learn and model the historical big data information including raw water quality indicators(such as raw water turbidity,water temperature,p H value,dissolved oxygen,oxygen consumption,etc.),influent flow and coagulant dosage,so as to predict coagulant dosage according to real-time raw water quality indicators and influent flow.Applying machine learning to coagulant dosing control and adjusting coagulant dosing in real time according to the current raw water quality index parameters is a subject worthy of further study:(1)In this paper,the raw water quality indicators(such as raw water turbidity,water temperature,p H value,dissolved oxygen,oxygen consumption,etc.),influent flow and coagulant dosage are taken as the research objects,and the historical operation big data are preprocessed,analyzed and modeled.Firstly,data cleaning,data dislocation processing and data normalization are carried out for historical operation big data;Secondly,five kinds of machine learning algorithms are used to analyze and model the raw water quality indexes(such as raw water turbidity,water temperature,p H value,dissolved oxygen,oxygen consumption)and influent flow as input and coagulant dosage as output;Finally,the coagulant dosage calculated by the data model is applied to the coagulant dosage control system.According to the actual effect of the effluent turbidity of the sedimentation tank,the random forest algorithm with the best precision among the five machine learning algorithms is selected.(2)In this paper,a long-term and short-term mixed model of coagulant dosing feedforward control based on random forest is proposed.The hybrid model includes long-term part and short-term part,in which the long-term part takes into account the periodicity of raw water quality changes,and carries out learning and data-driven modeling for the past five years’ historical operation data;In the shortterm part,the randomness of raw water quality change is considered,and the learning and data-driven modeling of historical operation data in recent 15 days are carried out;The weight of the long-term part and the short-term part is updated adaptively.The updating principle is to evaluate and adjust the weight of the long-term part and the short-term part according to the deviation between the effluent turbidity of the sedimentation tank and the set value.