Molecular descriptors-based models for estimating net heat of combustion of chemical compounds
Introduction
The heat of combustion () of a substance is the amount of heat, which is released from complete combustion at the oxygen-rich environment and it is normally defined at 298.15 K in 1 atm. Throughout the process, which includes oxidation to final products, the common products are CO2 (g), F2 (g), Cl2 (g), Br2 (g), I2 (g), SO2 (g), N2 (g), H3PO4 (s), H2O (g), and SiO2 (cristobalite) [1]. To produce heat or power, having from the fuel is vital to increase the efficiency of instruments. Measuring is essential to investigate the potential of fire threat of chemical compounds, differentiate the fuel heating value, materials stability, and engine performance such as emission and efficiency [[2], [3], [4], [5], [6], [7], [8]].
These values are gathered in some databases like API-TDB [9] and AIChE-DIPPR [1] for many chemicals. Because of time-consuming, being expensive experiments and even sometimes impossible procedure, determining by modeling methods would be faster, accurate, and more accessible. Using reliable relations of the determined could anticipate the heating value of other compounds and even un-synthesized materials. It is well known that the compound structure shows its property. Thus, well-established quantitative structure-property relationship (QSPR) technique has been extensively applied to estimate and correlate a variety of complicated and straightforward physicochemical properties of compounds in different areas. Until now, the property modeling has been investigated in few reports [[10], [11], [12], [13]] and the models work with combination of linear molecular descriptors as, exploring non-linear effects might surpass their predictive potential.
In the current study, new models were developed to forecast of many pure compounds, which belong to different chemical families. Moreover, the non-linear relationship between and molecular descriptors have been surveyed. Accordingly, Coupled Simulated Annealing optimization algorithm (CSA-LSSVM), Genetic Programming (GP) and Adaptive-Neuro Fuzzy Inference System optimized by Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) (PSO-ANFIS and GA-ANFIS) methods were employed for modeling. Table 1 shows brief explanations about the models developed in this study.
Section snippets
ANFIS
For the first time, Zade [14] presented the fuzzy logic (FL). The FL and ANN approaches are common in the Adaptive Neuro-Fuzzy Inference System (ANFIS). These two algorithms are coupled by explaining the If-then rules and certain functions called membership functions (MFs) for a construction called Fuzzy Inference (FIS). The modifications of MFs are based on the ability of the ANNs [[15], [16], [17]].
To optimize the ANFIS in the current study, two optimization algorithms were utilized. These
Data acquisition
In this study, data were collected from Refs. [11]: The database of the physical properties of pure chemicals of AIChE (American Institute of Chemical Engineers) is DIPPR 801 that was used in this work. In this database, 1714 data for pure compounds were found. Molecular descriptors are introduced as numerical features related to chemical structures. The molecular descriptor is the ultimate outcome of the rational and mathematical method that converts chemically encoded information in the
Model development
Simulated Annealing (SA) [[48], [49], [50]] is a precise algorithm to avoid local optima to expand local search methods. The basic approach leads to solve the lower quality than present solution to simplify the exit from local optimum. CSA, as a modification of SA, intends to far away from the local optima and advances the precision of solutions without reducing the rate of convergence, too much. Suykens et al. [51] displayed the method fundamentals that coupling in local optimization processes
Conclusion
In this work, CSA-LSSVM, GP, GA-ANFIS and PSO-ANFIS models with four and two inputs were developed for the estimation of . The parameters of LSSVM model were optimized to obtain an accurate model with the best performance using a simulated annealing algorithm. The ANFIS model was trained by PSO and GA algorithms to obtain high-accurate results and better performance. A comprehensive database (containing1714 points) of was used for model development. The data in the database was
Credit author statement
Amir Dashti: Conceptualization, Methodology, Software, Writing - original draft, Resources Omid Mazaheri.: Data curation, Formal analysis, Writing - review & editing. Farid Amirkhani: Visualization, Data curation, Investigation, Formal analysis. Amir H. Mohammadi: Supervision, Project administration, Conceptualization.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.None.
References (60)
Effects of combustion efficiency on a Dual cycle
Energy Convers Manag
(2009)- et al.
Multi-zone thermodynamic modelling of spark-ignition engine combustion–an overview
Energy Convers Manag
(2009) - et al.
Numerical analysis of methane–air combustion considering radiation effect
Energy Convers Manag
(2008) - et al.
Emission characterization and evaluation of natural gas-fueled cogeneration microturbines and internal combustion engines
Energy Convers Manag
(2008) - et al.
Investigating the effects of LPG on spark ignition engine combustion and performance
Energy Convers Manag
(2005) Examination of combustion characteristics and phasing strategies of a natural gas HCCI engine
Energy Convers Manag
(2005)- et al.
Combustion and exhaust emission characteristics of a dual fuel compression ignition engine operated with pilot diesel fuel and natural gas
Energy Convers Manag
(2004) A simple equation for prediction of net heat of combustion of pure chemicals
Chemometr Intell Lab Syst
(2008)- et al.
A new general method for estimation of heats of combustion for hazard evaluation
J Loss Prev Process Ind
(1990) Modeling discharge-suspended sediment relationship using least square support vector machine
J Hydrol
(2012)