Study Of Predictive Control In Water Treatment Coagulant Dosing

Decades of actual industrial experience show that it is difficult or even impossible to obtain the complex process of accurate model. Model predictive control is not demanding the precision of the model, which has significant advantage in modeling and robustness, and restrains the sensitivity of the algorithm for the model changed parameters. Thus model predictive control is widely used in complex industrial control.The coagulant dosing process is one of the most important parts in water treatment. Coagulant technique is applied to remove suspended particulates and harmful substances in raw water to achieve the purpose of purification. The outcome of coagulant dosing process not only affects the quality of the final effluent but also has a significant impact on the treatment efficiency of flowing treatment. Traditional mathematical modeling method can’t adapt to the sharp change of water quality. Yet the control effect of single model isn’t very ideal. Different kinds of water quality need different control models, therefore, this article proposed0-1combined model, according to the change of source water flow to switch model. Considered the water plant’s economic performance, a two-layered predictive control strategy is proposed for the coagulant dosing systems. The main work and results are as follows:1. The paper puts forward a0-1combinatorial optimization model of water treatment of coagulant dosage, which dynamic selects the appropriate model for the different water quality features, based on the current water quality stability, to achieve good control effect. When the water quality is relatively stable. It only takes the regression model, otherwise, adopts the back propagation neural networks(BPNN) model.0-1combined model can make different models match different water characteristics.2. The paper makes a comparison between regression model and back propagation neural networks (BPNN) model by way of simulation analysis using actual data in water plant, according to the established0-1combined model. Back propagation neural networks(BPNN) model can directly transit from historical data to the process control, which can effectively avoid the various process parameters. Regression model is superior to back propagation neural networks(BPNN) model when water quality is relatively stable.3. The paper proposes a two-layered predictive control strategy for the coagulant dosing systems. Model predictive controller(MPC) calculation is separated into steady-state and dynamic optimization, considering the water plant’s economic performance, through the objective function to reflect economic performance. Theoretical analysis and simulation results show that this method reduces operating costs and achieve maximum economic efficiency.