Modeling the adsorption of phenols and nitrophenols by activated carbon using genetic programming
Introduction
Currently, the exponential growth of global population and the manner in which industrial societies are developing, have given rise to serious environmental problems. For instance, the intensive development of the chemical industry, that has been required to meet the growing demand, has produced large scale pollution from coal processing (Song et al., 2015) and carbon emissions (Song and Zhou, 2015); global issues with significant economic, social and environmental implications (Lee et al., 2016, Jaromir and Varbanov, 2016). For these reasons, multidisciplinary research and collaboration between stakeholders should be enhanced to stimulate the transition toward a more sustainable society (Niesten et al., 2016).
Regarding our water supply, the excessive applications of phenolic compounds leads to serious repercussions. Water is widely used as a raw material in the process industries for a wide range of products. In particular, phenol and phenolic compounds are common contaminants found in effluents from the production of plastics, leather, paint, textile, paper, printing and other petrochemical processes (Altenor et al., 2009). The wastewaters from these industries are generating considerable amounts of phenolic pollutants (Kumar and Jena, 2016), leading to increasingly stringent environmental regulations on the discharge of effluents (Klemeš et al., 2011). The main reason for this is that these contaminants are extremely toxic, even if they are present in small concentrations their impact on health include protein degeneration, tissue erosion, paralysis of the central nervous system, vomiting, difficulty in swallowing, anorexia, liver and kidney damage, headache, fainting and various mental disturbances (Qadeer and Rehan, 2002, Srivastava et al., 2006, Takahashi et al., 1994). In fact, phenolic compounds are classified as high priority pollutants due to their carcinogenic effect on human health and their harmful effects on wild life (Altenor et al., 2009). Therefore, it is necessary to remove these phenolic compounds from industrial effluents before discharging them into the water stream.
The US Environmental Protection Agency (EPA) has instituted a regulation for lowering phenol concentration in the wastewater to less than 1.0 mg/dm3 (Tian et al., 2007). Therefore, some researchers have proposed a variety of methods for their removal from wastewater (Ukrainczyk and McBride, 1992). Among current proposals, the removal of phenols by adsorption technologies is regarded as one of the best methods because it does not require a high operating temperature and can be considered to be a relatively simple process from an implementation standpoint. Indeed, various adsorbent solids have been used to remove phenolic compounds from wastewater, including activated carbon (Nouri et al., 2002), silica (K1 et al., 2002), polymeric resins (Abburi, 2003), fly ash (Sarkar et al., 2005) and kaolinite (Barhoumi et al., 2003).
This work studies the use of activated carbon for the adsorption of phenols and nitrophenols. In particular, the goal of this paper is find models that capture the nonlinear relationships between several relevant operating conditions, such as the contaminant, initial concentration, pH and contact time, to predict the adsorption efficiency in an aqueous solution. Modeling phenol and nitrophenols adsorption onto activated carbon is very complex, it requires the solution of systems of equations that involve the radiant energy balance, the spatial distribution of the adsorbed radiation, mass transfer and the mechanisms of the adsorption transport problem (Altenor et al., 2009, Kumar and Jena, 2016). The studied adsorption process involves forces near the surface of the molecules, as well as the attractive and repulsive mobility forces between molecules of various types and shapes, which include the effects of molecular dissymmetry on properties of matter, such as evaporation, condensation and reflection. The process depends on several factors and phenomena, exhibiting a nonlinear behavior which is difficult to describe by linear mathematical models, such as those derived by linear regression methods. In similar problems, these considerations have lead researchers to turn towards non-linear data driven modeling methods (Nia et al., 2014, Karimi and Ghaedi, 2014, Dil et al., 2015, Ghaedi et al., 2015a, Ghaedi et al., 2015b).
Appropriate process models could be used to take corrective actions when the adsorption efficiency is predicted to be outside the required levels. The ability to take such actions could help reduce the environmental impact of the effluents when they are discharged into the water stream.
Therefore, the first main contribution of this work is the proposal of a relatively simple data-driven approach to derive models of the adsorption efficiency, based on a real-world experimental study. The derived models can help researchers and practitioners predict the effect that variations in the operating conditions might have on the adsorption process, without recurring to costly and time-consuming experimental tests. The second contribution of this paper relates to the specific algorithmic approach taken in this work, which is based on the use of genetic programming (GP) (Koza, 1992). Basically, we pose a supervised learning problem to automatically derive descriptive models of the adsorption efficiency and solve the problem by means of GP.
From the field of evolutionary computation, GP is similar to other machine learning techniques such as neural networks, but requires less a priori knowledge of the solution structure (Langdon and Poli, 2002, Poli et al., 2008, Garg and Lam, 2015). In this work, we compare two state-of-the-art GP methods recently proposed by the authors: (1) GP with local search optimization (GP-LS); and (2) neat-GP, a variant that searches for parsimonious solutions. The GP algorithms are also compared with standard regression techniques, including multivariate linear, quadratic and robust regression. Moreover, GP is also compared with more powerful regression techniques, namely Multivariate Adaptive Regression Splines (Friedman, 1991) and the Fast Function Extraction algorithm (McConaghy et al., 2011). We present a comprehensive numerical and statistical comparison of the results, showing that GP can be used to construct useful predictive models of the phenol adsorption efficiency by activated carbon, outperforming all other methods.
The remainder of this paper is organized as follows. Section 2 provides a detailed description of the process of adsorption of phenol and nitrophenols by activated carbon. Afterward, section 3 provides an overview of GP and describes our GP-based modeling algorithms, neat-GP and GP-LS. Section 4 presents our experimental work, describing how we derive a representative dataset of the adsorption process, the experimental setup and a discussion of the main results. Finally, Section 5 presents our conclusions and outlines future work.
Section snippets
Phenol and nitrophenol adsorption by activated carbon
The process of adsorption refers to the adhesion of molecules from a mixture in a gaseous or liquid state onto a solid surface. This process creates a film of the adsorbate on the surface of the adsorbent. The present work focuses on activated carbon as an absorbent, which has been widely used in industry for many years, due to its uniform porous structure and appropriate selective adsorptive (Ruthven, 1984, Speight, 1991).
The surface properties of activated carbon used in this work are
Genetic programming
GP is part of the larger research area known as evolutionary computation, which deals with the development of global search and optimization algorithms that are based on biological evolutionary theory (Koza, 1992, Langdon and Poli, 2002, Poli et al., 2008, Garg and Lam, 2015), distinguishing itself from other evolutionary algorithms (EAs) in several ways. GP searches for syntactic expressions that perform some form of computation, starting from a pool of solution candidates. The goal is to find
Experiments
As stated before, the goal of this work is to derive a a model that describes the relationship between operating variables and adsorption efficiency of phenol and nitrophenols onto activated carbon. In particular, we use a data-driven approach, where real experimental data is generated, recorded and then used to pose a supervised-learning problem. Afterward, GP (GP-LS and neat-GP) is used to solve a supervised learning problem. Performance is compared with several baseline and state-of-the art
Summary, conclusions and future work
The effective planning and management of industrial process that release contaminants into our ecosystem has become a primary concern over recent decades, since contamination and resource scarcity problems have led to a variety of impacts on our well being. However, achieving a reasonable and efficient management strategy is difficult since many conflicting factors have to be balanced, due to complexities of real world problems.
This paper presents the first application of GP to model the
Acknowledgments
This research was partially supported by CONACYT (México) Basic Science Research Projects no.169133 and no.178323, CONACYT Fronteras de la Ciencia 2015-2 Project No. 944, TecNM (México) Research Projects no. 5636.15-P and no. 5621.15-P, as well as by FP7-Marie Curie-IRSES 2013 European Commission program with project ACoBSEC with contract no. 612689. First and fifth authors are supported by CONACYT graduate scholarships, respectively no. 294213 and no. 332554.
References (54)
Adsorption of phenol and p-chlorophenol from their single and bisolute aqueous solutions on amberlite xad-16 resin
J. Hazard. Mater.
(2003)- et al.
Adsorption studies of methylene blue and phenol onto vetiver roots activated carbon prepared by chemical activation
J. Hazard. Mater.
(2009) - et al.
Coadsorption of alkylphenols and nonionic surfactants onto kaolinite
Colloids Surfaces A Physicochem. Eng. Aspects
(2003) - et al.
Prediction of energy performance of residential buildings: a genetic programming approach
Energy Build.
(2015) - et al.
Improving environmental sustainability by formulation of generalized power consumption models using an ensemble based multi-gene genetic programming approach
J. Clean. Prod.
(2015) - et al.
Artificial neural network-genetic algorithm based optimization for the adsorption of phenol red (pr) onto gold and titanium dioxide nanoparticles loaded on activated carbon
J. Ind. Eng. Chem.
(2015) - et al.
A hybrid artificial neural network and particle swarm optimization for prediction of removal of hazardous dye brilliant green from aqueous solution using zinc sulfide nanoparticle loaded on activated carbon
Spectrochim. Acta Part A Mol. Biomol. Spectrosc.
(2015) - et al.
Application of artificial neural network and genetic algorithm to modeling and optimization of removal of methylene blue using activated carbon
J. Ind. Eng. Chem.
(2014) - et al.
Removal of methylene blue and phenol onto prepared activated carbon from fox nutshell by chemical activation in batch and fixed-bed column
J. Clean. Prod.
(2016) - et al.
Sustaining the low-carbon emission development in asia and beyond: sustainable energy, water, transportation and low-carbon emission technology
J. Clean. Prod.
(10 March 2017)
Studies on pore systems in catalysts: V. the t method
J. Catal.
The analysis of surface and pores of activated carbons by the adsorption of various kinds of gases
Evolutionary-computer-assisted design of image operators that detect interest points using genetic programming
Image and Vision Computing
Comprehensive efficiency evaluation of coal enterprises from production and pollution treatment process
J. Clean. Prod.
Identification of epilepsy stages from {ECoG} using genetic programming classifiers
Computers in Biology and Medicine
Adsorptive removal of phenol by bagasse fly ash and activated carbon: equilibrium, kinetics and thermodynamics
Colloids Surfaces A Physicochem. Eng. Aspects
Variation of biodegradability of nitrogenous organic compounds by ozonation
Water Res.
Kinetics of the electrochemical oxidation of 2-nitrophenol and 4-nitrophenol studied by in situ uv spectroscopy and chemometrics
Electrochim. acta
Evolving estimators of the pointwise hölder exponent with genetic programming
Inf. Sci.
Neat genetic programming: controlling bloat naturally
Inf. Sci.
Comparative adsorption behavior between phenol and p-nitrophenol by na- and hdtma-clinoptilolite-rich tuff
Environ. Earth Sci.
Energy consumption forecasting using semantic-based genetic programming with local search optimizer
Intell. Neurosci.
Application of artificial neural network and response surface methodology for the removal of crystal violet by zinc oxide nanorods loaded on activate carbon: kinetics and equilibrium study
J. Taiwan Inst. Chem. Eng.
Adsorption Analysis: Equilibria and Kinetics
Deap: evolutionary algorithms made easy
J. Mach. Learn. Res.
Multivariate adaptive regression splines
Ann. Stat.
Method for calculation of effective pore size distribution in molecular sieve carbon
J. Chem. Eng. Jpn.
Cited by (30)
Computational intelligence for empirical modeling and optimization of methylene blue adsorption phenomena using available local zeolites and clay of Morocco
2022, Journal of Cleaner ProductionCitation Excerpt :The authors also applied sensitivity analysis to find the most powerful and the weak variables in this relationship. Finally, an optimization was conducted using the Particle Swarm Optimization (PSO) algorithm which yielded a required sorption efficiency of 98% (Z-Flores et al., 2017). also used genetic programming to model the adsorption of phenols and nitrophenols by activated carbon.
Compressive strength of concrete cylindrical columns confined with fabric-reinforced cementitious matrix composites under monotonic loading: Application of machine learning techniques
2022, StructuresCitation Excerpt :To this end, there is a specific number of individuals known as the “pool of candidate solutions” that would be able to generate the next generations. Each candidate solution is regarded as a tree structure consisting of operators and operands [60]. Afterward, the generated solution is compared with the target by determining the fitness value.
Alzheimer's disease diagnosis using genetic programming based on higher order spectra features[Formula presented]
2022, Machine Learning with ApplicationsThe AI-assisted removal and sensor-based detection of contaminants in the aquatic environment
2022, Artificial Intelligence and Data Science in Environmental Sensing