Real-time model predictive control of a wastewater treatment plant based on machine learning.

Two separate goals should be jointly pursued in wastewater treatment: nutrient removal and energy conservation. An efficient controller performance should cope with process uncertainties, seasonal variations and process nonlinearities. This paper describes the design and testing of a model predictive controller (MPC) based on neuro-fuzzy techniques that is capable of estimating the main process variables and providing the right amount of aeration to achieve an efficient and economical operation. This algorithm has been field tested on a large-scale municipal wastewater treatment plant of about 500,000 PE, with encouraging results in terms of better effluent quality and energy savings.

A fuzzy quality index for the environmental assessment of a restored wetland.

This paper describes the feasibility study for the restoration of agricultural land with a tendency to become waterlogged into a natural wetland, conceived to mitigate floods and to remove nutrients from the water drained from the cultivated plots. The wetland model, developed in aquatox, includes the nutrient dynamics both in the water and in the sediment, and the vegetation that is expected to develop as a consequence of flooding. The model inputs were synthesized from historical time series of rainfall and chemical data collected over the last decade. The model outputs are used to compute a synthetic fuzzy quality index (FQI) to assess the removal efficiency of the wetland. This FQI is based on three main variables describing the ecosystem quality: chlorophyll-a, dissolved oxygen and total suspended solids. This index has the merit of being simple enough to be immediately grasped by non-technical people, like managers and stakeholders, to whom the restoration project is proposed. The simulations, performed under five differing loading scenarios demonstrate the feasibility of this solution, which is robust enough to accommodate a 50% increase in either nitrogen, phosphorous or organic matter.

Real-time fault detection and isolation in biological wastewater treatment plants.

Automatic fault detection is becoming increasingly important in wastewater treatment plant operation, given the stringent treatment standards and the need to protect the investment costs from the potential damage of an unchecked fault propagating through the plant. This paper describes the development of a real-time Fault Detection and Isolation (FDI) system based on an adaptive Principal Component Analysis (PCA) algorithm, used to compare the current plant operation with a correct performance model based on a reference data set and the output of three ion-specific sensors (Hach-Lange gmbh, Düsseldorf, Germany): two Nitratax NOx UV sensors, in the denitrification tank and downstream of the oxidation tanks, where an Amtax ammonium-N sensor was also installed. The algorithm was initially developed in the Matlab environment and then ported into the LabView 8.20 (National Instruments, Austin, TX, USA) platform for real-time operation using a compact Field Point, a Programmable Automation Controller by National Instruments. The FDI was tested with a large set of operational data with 1 min sampling time from August 2007 through May 2008 from a full-scale plant. After describing the real-time version of the PCA algorithm, this was tested with nine months of operational data which were sequentially processes by the algorithm in order to simulate an on-line operation. The FDI performance was assessed by organizing the sequential data in two differing moving windows: a short-horizon window to test the response to single malfunctions and a longer time-horizon to simulate multiple unrepaired failures. In both cases the algorithm performance was very satisfactory, with a 100% failure detection in the short window case, which decreased to 84% in the long window setting. The short-window performance was very effective in isolating sensor failures and short duration disturbances such as spikes, whereas the long term horizon provided accurate detection of long-term drifts and proved robust enough to allow for some delay in failure recovery. The system robustness is based on the use of multiple statistics which proved instrumental in discriminating among the various causes of malfunctioning.

Intelligent monitoring system for long-term control of Sequencing Batch Reactors.

This paper discusses the application of artificial intelligence (AI) concepts to the monitoring of a lab-scale Sequencing Batch Reactor (SBR) treating nitrogen-rich wastewater (sanitary landfill leachate). The paper describes the implementation of a fuzzy inferential system to identify the correct switching sequence of the process and discusses the results obtained with six months of uninterrupted operation, during which the process conditions varied widely. The monitoring system proved capable of adjusting the process operation, in terms of phase length and external COD addition, to the varying environmental and loading conditions, with a percentage of correct phase recognition in excess of 95%. In addition, the monitoring system could be remotely operated through the internet via TCP/IP protocol.

Development of a open-vessel single-stage respirometer.

This paper describes the development and accuracy analysis of a single-stage respirometer which can be used both in the laboratory for wastewater characterization and in the plant as a process instrument. It is based on an accurate model of parasitic aeration, making the two-stage assumption unnecessary. Its operation is supervised by a real-time software, written in Lab View, managing the various measurement procedures and estimating the wastewater characteristics. Its accuracy is assessed through sensitivity and error propagation analysis, proving superior to the conventional model. A laboratory implementation of the instrument was tested with readily degradable substrate, yielding consistent and accurate respirograms.

Optimisation of sanitary landfill leachate treatment in a sequencing batch reactor.

A bench-scale SBR was operated for almost three years in an attempt to optimise the treatment of leachates generated in old landfill. The results of the first two years were used to design a monitoring and control system based on artificial intelligence concepts. Nitrogen removal was optimized via the nitrite shortcut. Nitrification and N removal were usually higher than 98% and 90%, respectively, whereas COD (of the leachate) removal was approximately 30-40%. The monitoring and control system was demonstrated capable of optimizing process operation, in terms of phase length and external COD addition, to the varying loading conditions. Using the control system developed, a significant improvement of the process was obtained: COD and N load were increased (HRT decrease) and a significant decrease (approximately 34%) of the ratio of COD added to N leachate content was observed.

Control of SBR switching by fuzzy pattern recognition.

The sequencing batch reactor (SBR) is a widely used process for biological removal of nutrients (nitrogen and phosphorus) from wastewater. It is based on the metabolism of specialised bacteria, which under alternate anaerobic/aerobic conditions uptake phosphorus and perform denitrification. Intermittent operation is normally operated on a fixed switching schedule with ample margin for possible inaccuracies, with the result that the process operation is highly inefficient. This paper proposes a switching strategy based on the indirect observation of process state through simple physico-chemical measurements and the use of an inferential engine to determine the most appropriate switching schedule. In this way the duration of each phase is limited to the time strictly necessary for the actual loading conditions. Experimental results show that the treatment cycle can be significantly shortened, with the results that more wastewater can be treated. The switching strategy is based on innovative data-processing techniques applied to simple process signals including pH, oxido-reduction potential (ORP) and dissolved oxygen (DO). They include wavelet filtering for signal denoising and fuzzy clustering for features extraction and decision-making. The formation of a knowledge-base and its adaptation during the operation are also discussed.

Estimating costs and benefits of advanced control for wastewater treatment plants--the MAgIC methodology.

This paper discusses a methodology to estimate the costs and benefits of advanced control for wastewater treatment plants. The methodology has been applied to four wastewater treatment plants, representing four standard types of plants built in Flanders, Belgium. The paper outlines the methodology and illustrated results from one of the four design cases. General results are shown and contrasted with full-scale experience. The methodology appears to give realistic results and will be used for further refinement of default control algorithms for certain types of plants. A preliminary analysis indicates that on-line control can become cost-effective for plant sizes above 50,000 population equivalents.

Accuracy analysis of a respirometer for activated sludge dynamic modelling.

The aim of the paper is to assess the experimental errors arising from the operation of a closed respirometer using autotrophic biomass. A closed, intermittent-flow device has been set-up for the measurement of oxygen uptake rate (OUR) and parameter calibration. After describing the device structure and operation, the factors affecting accuracy have been assessed. Inaccuracies may be caused by two groups of parameters: design parameters, including flow rate, volume, sampling time, numerical algorithm, sample injection and environmental parameters, concerning the physicochemical conditions of the experiment, such as unwanted oxygen transfer, pH, and the influence of sludge condition on "start-up" behaviour. It is shown to what extent each of them affects the final accuracy of the OUR measurement. In the second part of the paper, the respirometric data are used to calibrate a two-step nitrification model and their impact on the accuracy of the estimation of model parameters is assessed. Confidence limits are derived for the identifiable parameter combinations and the practical identifiability assessed with the aid of trajectory sensitivity analysis.

Fuzzy predictive control for nitrogen removal in biological wastewater treatment.

Whenever the carbon/nitrogen ratio of a domestic wastewater is too low, full denitrification is difficult to obtain and an additional source of organic carbon has to be provided. Since loading conditions may vary appreciably over the diurnal cycle, depending on the weather and sewage conditions, dosing should be controlled by an adaptive regulator to keep into account the time-varying process dynamics. A fuzzy predictive controller is proposed in this paper and its performance is tested through numerical simulations. The new aspects brought forward are the use of an improved model for denitrification, the use of benchmark (i.e. thoroughly tested and standardised) input files and the conclusion about regulator performance in overall plant performance, in terms of carbon saving and discharge compliance.

Implementation, study and calibration of a modified ASM2d for the simulation of SBR processes.

An enhanced process model for SBRs has been developed. Though the basic mechanism largely draws on the Activated Sludge Model n. 2d, its new features are the splitting of the nitrification stage in a two-step process, according to the well known Nitrosomonas-Nitrobacter oxidation sequence, and an improved XPAO dynamics, involved in the anaerobic/aerobic phosphorus removal process. The model was implemented through the DLL technique allowing complied C++ modules to be linked to an ordinary Simulink block diagram. The static sensitivity study revealed that if the parameter vector is partitioned into subsets of biologically related parameters and calibrated separately, the calibration procedure does not present particularly difficult aspects. Trajectory sensitivity showed also to which extent data collection could be optimised in order to improve calibration accuracy. The study of the shape of the error functional generated by parameters couples allows a much more effective calibration strategy.

A georeferenced river quality model.

Water quality models have reached a high degree of sophistication, but their weak side remains user interface and output georeferencing. The aim of this paper is to propose an interfacing procedure between two widespread but specialised programming environments: ArcVIEW as a Geographical Information System (GIS) and Matlab as a scientific programming tool for numerical analysis. The proposed solution is based on a Dynamic Data Exchange (DDE) between the two programs in order to operate a Matlab-based water quality model from within the GIS environment. It is shown how special GIS objects must be created and how they operate to achieve the goal of having quality data created by the model placed on a geographical map, together with other site features.