Performance evaluation of microbial fuel cell by artificial intelligence methods
Introduction
Microbial fuel cell (MFC) is an electrochemical device that converts the chemical energy in organic wastes into electricity by means of catalytic activities (reactions) of living microorganisms (Wei, Yuan, Cui, Han, & Shen, 2012). The phenomenon of generation of electricity from the contaminants in wastewater could provide environmental and economic benefits. Researchers have conducted several studies on power generation from the organic wastes in water. MFC technology has promising benefits but its practical implementation is limited due to low power generation, low energy efficiency as well as high material costs involved. According to recent developments in microorganisms, electrodes, operating conditions, matrix, ionic strength, different substrates and electrochemical characteristics, it was found that both microorganism as well as a biological factors has a significant impact on the power generation and thus responsible for the successful implementation of MFC (Logan et al., 2006).
Researchers pointed out that number of factors influence the power generation in MFC. One such factor is temperature which directly influences growth and reproduction of microorganism thus affecting intracellular and extracellular chemical process. Experimental studies are often conducted to determine the effect of various factors that influence the performance of MFC. The effect of temperature and anode media on the performance of MFC was studied by Min, Roman, and Angelidaki (2008). The maximum power density estimated at temperatures 30 °C and 22 °C was found to be 70 mW/m2 and 43 mW/m2 respectively. It was also noticed that at 15 °C, there was no power generation. The reason for this behavior of MFC can be that the catalytic activity of microorganism could have become inactive at 15 °C. Similar behavior of MFC was observed by Zhang and Shen (2006). It can be observed from this study that the microbial activity became inactive at 50 °C. Enhanced performance of the MFC is observed in the temperature range of 25–45 °C. Wei et al. (2012) studied the effect of temperature and hydraulic retention on the performance of a two-chambered MFC. Their studies showed that the MFC exhibited a rapid start-up for high substrate concentrations and was able to generate maximum power for long period of time.
It is also important to note that though experiments can satisfactorily determine the influence of various factors on the performance characteristics of fuel cell, limitations exists on the factors of cost and time required to perform the experiments. To overcome this limitation, several studies focused on developing mathematical models for estimating the performance of MFC (Choi et al., 2003, Picioreanu et al., 2010, Pinto et al., 2010, Pinto et al., 2012, Sousa and Gonzalez, 2005, Yao et al., 2004). These mathematical models (Vijayaraghavan and Wong, 2013a, Wong and Vijayaraghavan, 2012b) prove to be an effective tool for investigating the influence of various factors on MFC performance in addition to being cost effective and less time consuming (Oliveira, Simões, Melo, & Pinto, 2013). The mathematical models are generally formulated using differential and algebraic equations (Vijayaraghavan and Wong, 2013b, Wong and Vijayaraghavan, 2012a) and their derivation is based on the different phenomenon taking place inside the MFC. Although, these models provide accurate prediction, the formulation of these models requires a thorough knowledge on the functionality and the configuration of the MFC system.
Application of soft computing methods such as genetic programming (GP), artificial intelligence (AI), fuzzy logic and neural networks can be used as an alternative method for modeling complex physical non-linear systems such as a fuel cell system. These methods require input training data which can be obtained from the experimental data that is based on a specific design and operating condition of a fuel cell. Based on the input, the soft computing method can then be able to generate meaningful solutions for complicated problems (Castelli et al., 2013, Garg et al., 2013a, Garg et al., 2013b, Nazari, 2012, Peteiro-Barral et al., 2013).
Additionally, among the various soft computing methods described above, GP offers the advantage of a fast and cost-effective explicit formulation of a mathematical model based on multiple variables without any pre-definition of non-linear structure of the model (Al-Sahaf et al., 2012, Giot and Rosenberger, 2012, Kala, 2012, Kovacic and Brezocnik, 2003, Mabu et al., 2013, Mabu et al., 2012, Milfelner et al., 2005, Tsakonas, 2013, Tsakonas and Gabrys, 2012). The analytical model hence obtained can then be used by fuel cell developers to optimize the performance of their fuel cell based on specific operating conditions. It is to the best of authors’ knowledge that limited or no work exists on the application of soft computing methods on the performance prediction of an MFC system. Hence, in the present study, the performance characteristics of a MFC is modeled using three soft computing techniques viz., multi-gene genetic programming (MGGP), support vector regression (SVR) and artificial neural network (ANN). These methods are applied to model the operating voltage of MFC based on two input parameters (temperature and ferrous sulfate concentrations) at two operating conditions such as before start-up (BS) and after start-up (AS). An explicit formulation is derived for the output voltage as a function of time and each input parameter (temperature and ferrous sulfate concentrations) for BS and AS operation of MFC. Additionally, the performance of MGGP method to those of SVR and ANN has been compared. The remainder of this paper is organized as follows. In Section 2, the experimental details of the MFC are discussed in brief. In Section 3, three AI methods namely the MGGP, SVR and ANN are discussed and applied to data collected in Section 2. In Section 4, the performance of three AI methods is compared. Finally, Section 5 concludes with the recommendations for future work.
Section snippets
Data collection from microbial fuel cell (MFC)
The experimental data was collected from the literature and used to train and test different AI models (Wei et al., 2012). The data obtained from the experiment comprise of two input parameters namely, temperature and ferrous sulfate concentrations and output parameter voltage. The experimental description is provided as discussed in (Wei et al., 2012). Experiments were performed on two-chambered MFC comprising of two plexiglass bottles, which served as an anode and a cathode, each with an
Multi-gene genetic programming
This section gives an overview of the theory and principle of the genetic programming (GP) method followed by discussion on the multi-gene genetic programming method (MGGP). GP based on principle of Darwinian natural selection generates computer programs or models for solving regression problems. The extension of genetic algorithms (GAs) is GP. The only difference between GP and GA is that the solutions are represented by tree structures in GP while the solutions are represented in string form
Evaluation of models and its comparison
Models for each AI method were formulated for measuring the effect of each input parameter (temperature and ferrous sulfate concentration) on MFC Voltage at two operating conditions (BS and AS). Square of correlation coefficient (R) used to evaluate the performance of three AI methods MGGP, SVR and ANN is given bywhere Mi and Ai are predicted and actual values respectively, Mi and are the average values of predicted and actual
Conclusion and future work
The paper addresses the problems while modeling complex MFC systems using differential and algebraic equations. To counter these issues, three AI models namely, MGGP, SVR and ANN were proposed for modeling the voltage parameter of MFC system during BS and AS operating condition. The performance of these potential methods was compared. The results conclude that the MGGP model have shown better generalization ability than those of SVR and ANN models. Between SVR and ANN, ANN has shown better
References (44)
- et al.
Two-tier genetic programming: Towards raw pixel-based image classification
Expert Systems with Applications
(2012) - et al.
Analysis of a variable speed vapor compression system using artificial neural networks
Expert Systems with Applications
(2013) - et al.
Prediction of high performance concrete strength using genetic programming with geometric semantic genetic operators
Expert Systems with Applications
(2013) - et al.
Membrane fluidity sensoring microbial fuel cell
Bioelectrochemistry
(2003) - et al.
An end stage kidney disease predictor based on an artificial neural networks ensemble
Expert Systems with Applications
(2013) - et al.
Genetic programming for multibiometrics
Expert Systems with Applications
(2012) - et al.
A canonical correlation neural network for multicollinearity and functional data
Neural Networks
(2004) - et al.
Erratum: Evolving neural network using real coded genetic algorithm for permeability estimation of the reservoir
Expert Systems with Applications
(2013) Multi-robot path planning using co-evolutionary genetic programming
Expert Systems with Applications
(2012)- et al.
Parameter estimation of a pressure swing adsorption model for air separation using multi-objective optimisation and support vector regression model
Expert Systems with Applications
(2013)
A neural-network system for predicting machining behaviour
Journal of Materials Processing Technology
Enhanced decision making mechanism of rule-based genetic network programming for creating stock trading signals
Expert Systems with Applications
Adaptability analysis of genetic network programming with reinforcement learning in dynamically changing environments
Expert Systems with Applications
Optimization of self-organizing polynomial neural networks
Expert Systems with Applications
Genetic equation for the cutting force in ball-end milling
Journal of Materials Processing Technology
Prediction performance of PEM fuel cells by gene expression programming
International Journal of Hydrogen Energy
Overview on the developments of microbial fuel cells
Biochemical Engineering Journal
A comparative study of the scalability of a sensitivity-based learning algorithm for artificial neural networks
Expert Systems with Applications
A two-population bio-electrochemical model of a microbial fuel cell
Bioresource Technology
Optimizing energy productivity of microbial electrochemical cells
Journal of Process Control
Mathematical modeling of polymer electrolyte fuel cells
Journal of Power Sources
Local and global optimization for Takagi–Sugeno fuzzy system by memetic genetic programming
Expert Systems with Applications
Cited by (84)
Optimization techniques for electrochemical devices for hydrogen production and energy storage applications
2024, International Journal of Hydrogen EnergyApplication of artificial intelligence-based methods in bioelectrochemical systems: Recent progress and future perspectives
2023, Journal of Environmental ManagementAnalysis and prediction of microbial fuel cell behaviour using MLP and SVR
2023, Journal of the Taiwan Institute of Chemical EngineersRecent advances in artificial neural network research for modeling hydrogen production processes
2023, International Journal of Hydrogen EnergyWhen bioprocess engineering meets machine learning: A survey from the perspective of automated bioprocess development
2023, Biochemical Engineering Journal