Petia Georgieva

Adaptive Recurrent Neural network control of biological wastewater treatment

Three adaptive neural network control structures to regulate a biological wastewater treatment process are introduced: indirect, inverse model, and direct adaptive neural control. The objective is to keep the concentration of the recycled biomass proportional to the influent flow rate in the presence of periodically acting disturbances, process parameter variations, and measurement noise. This is achieved by the so‐called Jordan Canonical Recurrent Trainable Neural Network, which is a completely parallel and parametric neural structure, permitting the use of the obtained parameters, during the learning phase, directly for control system design. Comparative simulation results confirmed the applicability of the proposed control schemes.

On-line monitoring of a sugar crystallization process

The present paper reports a comparative evaluation of four multivariate statistical process control (SPC) techniques for the on-line monitoring of an industrial sugar crystallization process. The process itself is challenging since it is carried out in multiple phases and there exists strong non-linear and dynamic effects between the variables. The methods investigated include classical on-line univariate statistical process control, batch dynamic principal component analysis (BDPCA), moving window principal component analysis (MWPCA), batch observation level analysis (BOL) and time-varying state space modelling (TVSS). The study is focused on issues of on-line detection of changes in crystallization process operation, the early warning of process malfunctions and potential production failures; problems that have not been directly addressed by existing statistical monitoring schemes. The results obtained demonstrate the superior performance of the TVSS approach to successfully detect abnormal events and periods of bad operation early enough to allow bad batches and related losses in amounts of recycled sucrose to be significantly reduced.

Knowledge-based hybrid modelling of a batch crystallisation when accounting for nucleation, growth and agglomeration phenomena

Abstract This paper reports on the application of knowledge-based hybrid (KBH) modelling to an industrial scale (fed-) batch evaporative crystallisation process in cane sugar refining. First, principles models of the process lead in general to good description of process state, except for the prediction of the main crystal size distribution (CSD) parameters—mean size and the coefficient of variation. This is due to difficulties in expressing accurately nucleation and crystal growth rates and especially the complex phenomena of agglomeration in the relevant population balance. A hybrid model is proposed, which combines a partial mechanistic model that reflects the general mass, energy and population balances with a neural network to express growth rate, nucleation kinetics and agglomeration phenomena. Results obtained demonstrate a better agreement between experimental data and hybrid model predictions than that observed with the complete mechanistic model.

Adaptive k-tracking control of activated sludge processes

An adaptive controller for activated sludge processes is introduced. The control objective is to keep, in the presence of input constraints, the concentration of the biomass proportional to the in ̄ uent ̄ ow rate, where a prespeci® ed small tracking error of size ¶ is tolerated. This is achieved by the so called ¶-tracker which is simple in its design, relies only on structural properties of the process and weak feasibility properties, and does not invoke any estimation or identi® cation mechanism or probing signals. ¶-Tracking is proved for a model of an activated sludge process with unknown reaction kinetics and including unknown time-varying process parameters. It is illustrated by simulations that the ¶-tracker works successfully, and even under practical circumstances which go beyond what we can prove mathematically, it can cope with `white noise’ corrupting the measurement and periodically acting disturbances.

Robust control design of an activated sludge process

SUMMARY A H-inf control strategy is presented for a robustly performing activated sludge process. In operational terms, the objective is to conduct the process imposing that the biomass concentration in the recycle stream follows closely a time-varying, process-dependent, reference signal. The corresponding control objective is described as the development of a robust reference-tracking control structure with the best possible disturbance compensation, able to perform with noisy measurements and able to cope with variations in key process model parameters. The proposed algorithm embeds an estimation of states by solving the Riccati equation and avoids parameter estimation by assuming their variability within known bounds. Results are presented which show a favourable behaviour of the robust controller in comparison with a conventional PI control structure. Copyright ( 1999 John Wiley & Sons, Ltd.

Advances in EEG-Based biometry

This paper is focused on proving the concept that the EEG signals collected during a perception or mental task can be used for discrimination of individuals. The viability of the EEG-based person identification was successfully tested for a data base of 13 persons. Among various classifiers tested, Support Vector Machine (SVM) with Radial Basis Function (RBF) exhibits the best performance. The problem of static classification that does not take into account the temporal nature of the EEG sequence was considered by an empirical post classifier procedure. The algorithm proposed has an effect of introducing a memory into the classifier without increasing its complexity. Control of a classified access into restricted areas security systems, health disorder identification in medicine, gaining more understanding of the cognitive human brain processes in neuroscience are some of the potential applications of EEG-based biometry.

Dynamic Optimisation of Industrial Sugar Crystallization Process based on a Hybrid (mechanistic+ANN) Model

A model-based optimization of an industrial fed-batch sugar crystallisation process is considered in this paper. The objective is to define the optimal profiles of the manipulated process inputs, the feeding rate of liquor/syrup and the steam supply rate, such that the crystal content and the crystal size distribution (CSD) measures at the end of the batch cycle reach the reference values. A knowledge-based hybrid model is implemented, which combines a partial first principles model reflecting the mass, energy and population balances with an artificial neural network (ANN) to estimate the kinetics parameters - particle growth rate, nucleation rate and the agglomeration kernel. The simulation results demonstrate that the very tight and conflicting end-point objectives are simultaneously feasible in the presence of hard process constrains.

Modelling and Adaptive Control of Aerobic Continuous Stirred Tank Reactors

A biotechnological aerobic process is modelled as an ordinary differential equation which, under mild assumptions, ensures invariance of the positive orthant and boundedness of the concentrations. An adaptive controller is designed for this general class of processes so that the external substrate can be regulated by the dilution rate into a prespecified arbitrarily small neighbourhood of a constant setpoint reference. The adaptive controller is robust, simple in its design without invoking any identification mechanisms, and is based on output data only. It is shown that the prominent example of a bakers yeast fermentation belongs to this setup, and adaptive tracking is illustrated by simulations.

Sparse Nonnegative Matrix Factorization with Genetic Algorithms for Microarray Analysis

Nonnegative Matrix Factorization (NMF) has proven to be a useful tool for the analysis of nonnegative multivariate data. Gene expression profiles naturally conform to assumptions about data formats raised by NMF. However, it is known not to lead to unique results concerning the component signals extracted. In this paper we consider an extension of the NMF algorithm which provides unique solutions whenever the underlying component signals are sufficiently sparse. A new sparseness measure is proposed most appropriate to suitably transformed gene expression profiles. The resulting fitness function is discontinuous and exhibits many local minima, hence we use a genetic algorithm for its optimization. The algorithm is applied to toy data to investigate its properties as well as to a microarray data set related to Pseudo-Xanthoma Elasticum (PXE).

EEG Signal Processing for Brain–Computer Interfaces

This chapter is focused on recent advances in electroencephalogram (EEG) signal processing for brain computer interface (BCI) design. A general overview of BCI technologies is first presented, and then the protocol for motor imagery noninvasive BCI for mobile robot control is discussed. Our ongoing research on noninvasive BCI design based not on recorded EEG but on the brain sources that originated the EEG signal is also introduced. We propose a solution to EEG-based brain source recovering by combining two techniques, a sequential Monte Carlo method for source localization and spatial filtering by beamforming for the respective source signal estimation. The EEG inverse problem is previously studded assuming that the source localization is known. In this work for the first time the problem of inverse modeling is solved simultaneously with the problem of the respective source space localization.