Optimization Of Coagulant Dosage Based On Artificial Intelligence

With the rapid development of artificial intelligence and the continuous improvement of labor costs,people have also put forward higher requirements for life water quality.Intelligent and automated management has become the future application trend of drinking water plants.As the front end of the treatment process of drinking water plant,coagulation process has an important impact on the quality of subsequent effluent water,and controlling the precise dosage of coagulant is of great significance to the water quality requirements and operating costs of the water plant.There is a complex nonlinear relationship between coagulant dosage and effluent quality,so it is difficult to control its precise dosage.At present,artificial intelligence technology has strong nonlinear fitting ability,and research shows that there is a high correlation between floc image characteristics and effluent water quality.Therefore,this study is based on artificial intelligence correlation algorithm to study floc image combined with water quality characteristics to predict coagulant dosage.Based on the beaker test,five kinds of artificial simulated water samples were configured in this study,and four groups of water quality parameters and floc images were collected for each group of water samples.Two kinds of coagulant dosage prediction models were established according to floc images and water quality parameters,and their prediction effects were discussed as follows:(1)Build a set of floc image acquisition device and water quality parameter acquisition device,which can realize continuous collection of high-quality floc images,and can continuously collect relevant water quality parameters(turbidity,COD,temperature,pH,UV254 and A520)in the coagulation process,meeting the data set required for model training.(2)Four different deep learning models(AlexNet,VggNet,GoogleNet and ResNet)were successfully established to predict the dosage of coagulant.By adding two kinds of coagulants with different concentrations and choosing beakers to test two kinds of floc images with different simultaneous lengths,there are 16 kinds of models in total.The best model is the GoogleNet model with the dosage of 20mg/L coagulant and the input image is the whole process image,with the highest prediction accuracy of 98.7%.The worst model is ResNet50 model based on transfer learning,and the highest prediction accuracy is the whole process flocculation image with 10mg/L dosage,and its verification accuracy is 92.2%.Then VGGNet16(PAC=20mg/L;1920s)model with the highest validation accuracy of 98%,AlexNet(PAC=20mg/L;1920s)model,the highest validation accuracy was 92.7%.GoogleNet(PAC=20mg/L;The 1920s model was evaluated,and the results showed that the model could correctly classify most of the floc images in any of the five types of floc images and had a certain generalization ability.(3)The Opencv algorithm was used to process the image gray-scale,simplifying the complexity of the image.Then 170 was selected as the segmentation threshold through the preexperiment.Subsequently,four different floc image feature parameters were successfully obtained through the image segmentation technology.The results show that the number of floc,particle size distribution characteristics,floc image density,fractal dimension and their variation rules are consistent with the mechanism of coagulation process,and can be used as parameters to describe the coagulation process under the study conditions,and can be used as training data for subsequent artificial neural network models.(4)Based on the bp neural network model,the flocculant characteristics(number of flocs,fractal dimension,flocs density and particle size distribution)and water quality characteristics(turbidity,COD,temperature,pH,UV254 and A520 absorbance)were used as the input of the model to predict the optimal dosage of coagulant.The experimental results showed that the accuracy of the model was 97.5%.We also use some model evaluation methods,and the results show that the model has high reliability.