Intelligent Dosing Method for Water Treatment Based on Data Mining Combined with Image Recognition and Self-Learning

Abstract

To deal with the low efficiency in the traditional flocculation link of steel plant wastewater treatment, an intelligent dosing system combined with image recognition, LightGBM algorithm, and self-learning ability is proposed to automatically monitor water quality and recommend optimal dosage analyzing both image and water quality parameters. Images acquired from the industrial camera and water parameters from underwater detectors are used as input for the dosing models which are trained by the LightGBM algorithm, then the recommended dosage could be given by the model and sent to an automatic dosing system controlled by a programmable logic controller (PLC), which forms a closed- loop for intelligent dosing. The resulting parameters are also iteratively recorded for self-learning. Based on two high-density pools with one adopting the proposed intelligent dosing system and the other completely manually-controlled, the experiment shows that the proposed intelligent dosing system completely realizes unmanned control and improves the average water qualification rate up to 97.52%, increasing it by 85.81% compared with traditional artificial dosing, and dosing costs decrease by approximately 41%.