Yoon-Seok Timothy Hong

Analysis of a municipal wastewater treatment plant using a neural network-based pattern analysis

This paper addresses the problem of how to capture the complex relationships that exist between process variables and to diagnose the dynamic behaviour of a municipal wastewater treatment plant (WTP). Due to the complex biological reaction mechanisms, the highly time-varying, and multivariable aspects of the real WTP, the diagnosis of the WTP are still difficult in practice. The application of intelligent techniques, which can analyse the multi-dimensional process data using a sophisticated visualisation technique, can be useful for analysing and diagnosing the activated-sludge WTP. In this paper, the Kohonen Self-Organising Feature Maps (KSOFM) neural network is applied to analyse the multi-dimensional process data, and to diagnose the inter-relationship of the process variables in a real activated-sludge WTP. By using component planes, some detailed local relationships between the process variables, e.g., responses of the process variables under different operating conditions, as well as the global information is discovered. The operating condition and the inter-relationship among the process variables in the WTP have been diagnosed and extracted by the information obtained from the clustering analysis of the maps. It is concluded that the KSOFM technique provides an effective analysing and diagnosing tool to understand the system behaviour and to extract knowledge contained in multi-dimensional data of a large-scale WTP.

Intelligent characterisation and diagnosis of the groundwater quality in an urban fractured-rock aquifer using an artificial neural network

Abstract This paper addresses the problem of how to diagnose the effect of stormwater infiltration on groundwater quality variables and to capture the complex nonlinear relationships that exist between groundwater quality variables. It is argued that because of the complex nonlinear relationships between the groundwater quality variables, classical linear statistical methods are unreliable and difficult to visualise the results. The application of intelligent techniques, which can analyse the multi-dimensional groundwater quality data with the sophisticated visualisation technique, is vital for sustainable groundwater management. In this paper, the Kohonen self-organising feature maps (KSOFM) neural network is applied to analyse the effect of stormwater infiltration on the groundwater quality, and diagnose the inter-relationship of the groundwater quality variables in a fractured rock aquifer. Based on the pattern analysis visualised in component planes and U-matrix, the inter-relationships among the groundwater quality variables due to the stormwater infiltration are extracted and interpreted. The pattern distribution of groundwater quality variables due to different aquifer conditions is also analysed. It is concluded that the KSOFM technique described in this paper provides an effective analysing and diagnosing tool to understand the dynamic in the groundwater quality and to extract knowledge contained in the multi-dimensional data. Finally it has considerable potential not only in groundwater quality monitoring and diagnosis, but also in other environmental areas.

Identification of an urban fractured-rock aquifer dynamics using an evolutionary self-organizing modelling

Abstract An urban fractured-rock aquifer system, where disposal of storm water is via ‘soak holes’ drilled directly into the top of fractured-rock basalt, has a highly dynamic nature where theories or knowledge to generate the model are still incomplete and insufficient. Therefore, formulating an accurate mechanistic model, usually based on first principles (physical and chemical laws, mass balance, and diffusion and transport, etc.), requires time- and money-consuming tasks. Instead of a human developing the mechanistic-based model, this paper presents an approach to automatic model evolution in genetic programming (GP) to model dynamic behaviour of groundwater level fluctuations affected by storm water infiltration. This GP evolves mathematical models automatically that have an understandable structure using function tree representation by methods of natural selection (‘survival of the fittest’) through genetic operators (reproduction, crossover, and mutation). The simulation results have shown that GP is not only capable of predicting the groundwater level fluctuation due to storm water infiltration but also provides insight into the dynamic behaviour of a partially known urban fractured-rock aquifer system by allowing knowledge extraction of the evolved models. Our results show that GP can work as a cost-effective modelling tool, enabling us to create prototype models quickly and inexpensively and assists us in developing accurate models in less time, even if we have limited experience and incomplete knowledge for an urban fractured-rock aquifer system affected by storm water infiltration.

Sequential modeling of fecal coliform removals in a full-scale activated-sludge wastewater treatment plant using an evolutionary process model induction system

We propose an evolutionary process model induction system that is based on the grammar-based genetic programming to automatically discover multivariate dynamic inference models that are able to predict fecal coliform bacteria removals using common process variables instead of directly measuring fecal coliform bacteria concentration in a full-scale municipal activated-sludge wastewater treatment plant. A sequential modeling paradigm is also proposed to derive multivariate dynamic models of fecal coliform removals in the evolutionary process model induction system. It is composed of two parts, the process estimator and the process predictor. The process estimator acts as an intelligent software sensor to achieve a good estimation of fecal coliform bacteria concentration in the influent. Then the process predictor yields sequential prediction of the effluent fecal coliform bacteria concentration based on the estimated fecal coliform bacteria concentration in the influent from the process estimator with other process variables. The results show that the evolutionary process model induction system with a sequential modeling paradigm has successfully evolved multivariate dynamic models of fecal coliform removals in the form of explicit mathematical formulas with high levels of accuracy and good generalization. The evolutionary process model induction system with sequential modeling paradigm proposed here provides a good alternative to develop cost-effective dynamic process models for a full-scale wastewater treatment plant and is readily applicable to a variety of other complex treatment processes.

Inference model derivation with a pattern analysis for predicting the risk of microbial pollution in a sewer system

Developing a mathematical model for predicting fecal coliform bacteria concentration is very important because it can provide a basis for water quality management decisions that can minimize microbial pollution risk to the public. This paper introduces a hybrid modeling methodology which is a combined use of a neural network-based pattern analysis and an evolutionary process model induction system. The neural network-based pattern analysis technique is applied to extract knowledge on inter-relationships between fecal coliform concentrations and other measurable variables in a sewer system. Based on the result of neural network-based pattern analysis, an evolutionary process model induction system is used to derive mathematical inference models that can predict fecal coliform bacteria concentration from easily measurable variables instead of directly measuring fecal coliform bacteria concentration in a sewer system. The neural network-based pattern analysis extracts that temperature and ammonia concentration are the most important driving forces leading to an increase in fecal coliform bacteria concentration in the sewer system at Paraparaumu City, New Zealand. Fecal coliform bacteria concentration is also positively correlated with dissolved phosphorus and inversely with flow rate. The multivariate inference models that are able to predict fecal coliform bacteria concentration are successfully derived as functions of flow rate, temperature, ammonia, and dissolved phosphorus in the form of understandable mathematical formulae using the evolutionary process model induction system, even if a priori mathematical knowledge of the dynamic nature of fecal coliform bacteria is poor. The multivariate inference models evolved by the evolutionary process model induction system produce a slightly better performance than the multi-layer perceptron neural network model.

Online nonlinear sequential Bayesian estimation of a biological wastewater treatment process

Online estimation of unknown state variables is a key component in the accurate modelling of biological wastewater treatment processes due to a lack of reliable online measurement systems. The extended Kalman filter (EKF) algorithm has been widely applied for wastewater treatment processes. However, the series approximations in the EKF algorithm are not valid, because biological wastewater treatment processes are highly nonlinear with a time-varying characteristic. This work proposes an alternative online estimation approach using the sequential Monte Carlo (SMC) methods for recursive online state estimation of a biological sequencing batch reactor for wastewater treatment. SMC is an algorithm that makes it possible to recursively construct the posterior probability density of the state variables, with respect to all available measurements, through a random exploration of the states by entities called ‘particle’. In this work, the simplified and modified Activated Sludge Model No. 3 with nonlinear biological kinetic models is used as a process model and formulated in a dynamic state-space model applied to the SMC method. The performance of the SMC method for online state estimation applied to a biological sequencing batch reactor with online and offline measured data is encouraging. The results indicate that the SMC method could emerge as a powerful tool for solving online state and parameter estimation problems without any model linearization or restrictive assumptions pertaining to the type of nonlinear models for biological wastewater treatment processes.