Soft Sensors Of Water Quality For Wastewater Treatment Plant Of Activated Sludge Process

Wastewater treatment process is a significant process to remove carbonaceous, nitrogenous, phosphorous organisms and poisonous heavy metals from municipal and industrial wastewater, which is aimed at improving water qualities under emission limits. Especially, COD (Chemical Oxygen Demand) becomes the key control index of total contamination since it is a crucial water quality index indicating organic pollution.Since effluent COD of middle working procedures anoxic and aerbic tank as well as secondary sedimentation tank in A/O (Anoxic/Oxic) process cannot be measured online, most measurement methods for COD are off-line which have the shortcomings of complex operation, large time delay, it is difficult to realize real-time feedback control of water qualities promptly, so soft sensors of water quality are needed. The biological mechanisms of wastewater treatment process are complex; operating conditions are fluctuated frequently with influent load. Wastewater treatment process has strong nonlinearity, uncertainty and time-varying property caused by production conditions, influent qualities and biological reactions which is difficult to model wastewater treatment process accurately. Soft sensors of water quality based on pure mechanism model are difficult to guarantee accuracy of soft sensor. In order to monitor effluent qualities online, guarantee effluent indices and enhance contamination removal efficiency, precise soft sensors of water quality which can estimate water qualities promptly are needed.On-line soft sensors of effluent water quality COD are investigated in this thesis. The main contributions of this thesis are listed as follows:(1) Aiming at the problems that neural networks modeling methods with general error backpropagation are difficult to guarantee the stability of modeling error when nonlinear system bears unmodeled dynamics and uncertain disturbances, a wavelet neural network(WNN) modeling method based on stable learning law is presented by which modeling error is guaranteed. Stable learning law is deduced based on ISS(Input-to-State Stability) theory, and the stability of modeling error of wavelet neural model is analyzed. Capabilities between WNN model based on stable learning law and general error backpropagation algorithm with and without noise respectively, WNN based on stable learning law and MLP based on general error backpropagation algorithm are compared. The presented WNN is used to model soft sensor of effluent COD from secondary sedimentation tank. The simulation is conducted using real data from wastewater treatment plant in Shenyang, and the results show the validity of WNN modeling method based on stable learning law.(2) Aiming at the problems that effluent quality from secondary sedimentation tank is difficult to be measured online, and the precision of water quality soft sensor is deteriorated by fluctuating operating conditions with high, normal and low load, an on-line COD soft sensor based on multi-models integrating WNN and Hammerstein model(for short H model) is proposed. Soft sensor model is composed of operating condition recognition mechanism, local model based on H model under each operating condition, multi-models combination mechanism. Among these, operating condition recognition mechanism adopting subtractive clustering is used to partition operating condition according to high, normal and low load, and to ascertain the locations of initial operating centers. Operating centers are corrected by similarity function between new characteristic variables of operating conditions and operating centers, and the subsidiary values of new characteristic variables to corrected operating centers are calculated. Local model based on H model is built within each operating condition where the nonlinear part and linear part of H model are modeled by WNN based on stable learning law and ARX model based on identification algorithm of recursive least square(RLS), respectively. Multi-models combination mechanism is based on weighted sum of the outputs from local H models to calculate the output of COD soft sensor with the subsidiary values of new characteristic variables to corrected operating centers as weight. The simulation using real operating data shows that the on-line multi-models COD soft sensor with correction of operating centers performs high precision even under the fluctuations of varying operating conditions.(3) Aiming at the problems that outlet qualities of anoxic and aerbic tank as well as secondary sedimentation tank in A/O process cannot be measured online, and analyzed water qualities values of anoxic and aerobic tank are lacked, data-driven soft sensors are not applicable to estimate effluent qualities of middle working procedures effectively, an on-line hybrid COD soft sensor based on mechanism model of activated sludge process and neural network is presented. Hybrid soft sensor is composed of anoxic and aerobic tank models based on mechanism model, secondary sedimentation tank model, anoxic and aerobic reaction rate models based on MLP, component calculation model, water quality calculation model. Among these, anoxic and aerobic tank models adopt SASM1(Simplified Activated Sludge Model No.1) reduced from ASM1. Secondary sedimentation tank model adopts ideal sedimentation model. Anoxic and aerobic reaction rate models adopt MLP to identify reaction rates of sensitive components in anoxic and aerobic tank, in which the relationships between modeling error of hybrid model and those of anoxic reaction rate models as well as aerobic reaction rate models are obtained through error backpropogation and chain rule, by which anoxic reaction rate model and aerobic reaction rate model are learned respectively. Component calculation model divides influent water quality to components in SASM1by percentage. Water quality calculation model converts component concentration to water quality COD. The simulations using real data show that on-line hybrid COD soft sensor performs high performance and can estimat effluent qualities of anoxic and aerobic tank as well as secondary sedimentation tank at the same time.