Article pubs.acs.org/IECR
Artificial Intelligence-based Modeling of High Ash Coal Gasification in a Pilot Plant Scale Fluidized Bed Gasifier Veena Patil-Shinde,† Tejas Kulkarni,†,∥ Rahul Kulkarni,† Prakash D. Chavan,‡ Tripurari Sharma,§ Bijay Kumar Sharma,‡ Sanjeev S. Tambe,*,† and B. D. Kulkarni*,† †
Chemical Engineering and Process Development Division, CSIR-National Chemical Laboratory, Dr. Homi Bhabha Road, Pashan, Pune 411008, India ‡ CSIR-Central Institute of Mining and Fuel Research (CIMFR), Dhanbad, Jharkhand 828108, India § Indian School of Mines (ISM), Dhanbad, Jharkhand 826004, India S Supporting Information *
ABSTRACT: The quality of coalespecially its high ash contentsignificantly affects the performance of coal-based processes. Coal gasification is a cleaner and an efficient alternative to the coal combustion for producing the syngas. The high-ash coals are found in a number of countries, and they form an important source for the gasification. Accordingly, in this study, extensive gasification experiments were conducted in a pilot-plant scale fluidized-bed coal gasifier (FBCG) using high-ash coals from India. Specifically, the effects of eight coal and gasifier process related parameters on the four gasification performance variables, namely CO+H2 generation rate, syngas production rate, carbon conversion, and heating value of the syngas, were rigorously studied. The data collected from these experiments were used in the FBCG modeling, which was conducted by utilizing two artificial intelligence (AI) strategies namely genetic programming (GP) and artificial neural networks (ANNs). The novelty of the GP formalism is that it searches and optimizes both the form and parameters of an appropriate linear/nonlinear function that best fits the given process data. The original eight-dimensional input space of the FBCG models was reduced to three-dimensional space using the principal component analysis (PCA) and the PCA-transformed three variables were used in the AI-based FBCG modeling. A comparison of the GP and ANN-based models reveals that their output prediction accuracies and the generalization performance vary from good to excellent as indicated by the high training and test set correlation coefficient magnitudes lying between 0.92 and 0.996. This study also presents results of the sensitivity analysis performed to identify those coal and process related parameters, which significantly affect the FBCG process performance.
1. INTRODUCTION The widely employed coal-based combustion technologies for the power generation suffer from a number of drawbacks, such as lower efficiency and significant emissions of CO2, SOx, and NOx gases. These emissions lead to the climate change and air pollution. An important factor that influences the performance of a thermal power station is the quality of the coal used in the combustion. Specifically, the usage of high ash (>20%) coal produces the following adverse effects: (a) emission of more particulate matter into the atmosphere, (b) reduced power station boiler efficiency, (c) lower boiler efficiency leads to consuming higher volumes of coal to achieve the targeted power output, which results in higher coal transportation costs and, consequently, costlier power, and (d) higher levels of impurities from the coal (e.g., ash and moisture) that do not contribute to the combustion process; these also lead to severe waste disposal problems. There are a number of countries such as India, China, Australia, and Turkey, where high ash coal deposits are found. In India, coal-based energy meets nearly 70% of the country’s energy needs. The Indian thermal power stations invariably receive high ash coals, and therefore, CO2 emission control has become a major concern. For achieving the stringent pollution control targets, changes in the coal utilization practices and the development of clean coal technologies have become essential. © 2014 American Chemical Society
These measures are expected to result in a high coal conversion efficiency and lower environmental impact.1 The gasification of coal is such a promising clean coal technology.2 The typical thermal efficiencies of the conventional pulverized-fuel (PF)fired power stations are approaching 37%, whereas supercritical PF units can achieve net efficiencies of 47%.3 In comparison, power generation using an integrated gasification combinedcycle (IGCC) system has achieved thermal efficiencies of approximately 47%,4 and it is believed that the efficiencies exceeding 50% are possible in the near future.3,5 The newer gas turbine concepts and increased process temperatures are targeting efficiencies up to 65%.5 The gasification technology, being environment-friendly, is a potential alternative to the conventional coal combustion-based power generation. The conventional thermal power plants with steam cycles alone cannot achieve the high efficiency targets, and hydrogen production from the combustion plants is not feasible. These limitations are not applicable to the gasification technologies and they possess several other advantages as well Special Issue: Ganapati D. Yadav Festschrift Received: Revised: Accepted: Published: 18678
February 13, 2014 April 8, 2014 April 9, 2014 April 9, 2014 dx.doi.org/10.1021/ie500593j | Ind. Eng. Chem. Res. 2014, 53, 18678−18689
Industrial & Engineering Chemistry Research
Article
due to their flexibility in the syngas applications. There exist three major coal gasification technologies, namely moving (fixed) bed, fluidized bed, and entrained bed gasifiers. Among these, the fluidized bed coal gasifier (FBCG) possesses the following advantages:6 • Process conditions in an FBCG are more uniform leading to the better heat and mass transfer in the bed and steady product composition. • Provides better contact between the solid and gaseous reactants, which is favorable for maximizing the carbon conversion. • It has a high solids residence time, can use bigger particle sizes, and is capable of handling the high-sulfur coals without a need for the flue-gas desulfurization systems, which incur high capital and operational costs. • It operates at lower temperatures and thus emits lower amounts of nitrogen oxides (NOx). The low temperature process, besides improving the system’s reliability, is also inherently more energy efficient since it utilizes nearly all the heat generated in the gasifier to support the gasification. • The FBCG can handle a wide variety of coals ranging from the high quality bituminous coal to the lignite. For more reactive fuels, such as the sub-bituminous coal and lignite, the fluidized bed gasifiers can achieve good gasification yields and carbon conversion at relatively mild conditions. • The amount of tar and phenol formation is low or negligible. • The large fuel inventory provides safety, reliability, and stability. • Potential for in situ sulfur capture. • Better turn-down ratio. Owing to the above-stated several attractive characteristics, the fluidized bed coal gasifiers are better suitedin comparison with the other types of gasifiersfor handling high ash coals. A large number of studies have been performed to investigate the fluidized bed coal gasification (see for example Gutierrez and Watkinson,7 Ocampo et al.,8 and Ju et al.9). However, a detailed experimental investigation and analysis of the fluidized bed gasification with a focus on the high ash coals is not found in the open literature. Such a study is important since in countries like India a large portion of the energy comes from the coal-fired thermal power stations and there exists a dire need to switch over to more efficient clean coal technologies, such as the gasification. Another reason for studying the FBCG in depth is the following: the gasifier operates in the dry ash removal mode owing to which the operating temperature needs to be lower. This may result in an unconverted carbon in the fly and bottom ashes. To address the issues arising from the low temperature FBCG operation, an in-depth investigation of the various factors affecting the FBCG performance has become necessary. Accordingly, this study first reports the results of the fluidized bed gasification in a pilot-plant scale gasifier using high ash coals of Indian origin. Availability of an accurate, robust, and reliable mathematical process model of an FBCG assists in the preliminary process design, complex simulation, prediction of the steady-state and dynamic behavior, startup, shutdown, change of fuel and load, scaling up, control, fault detection and diagnosis, and process optimization. Such models are also helpful in fixing the right magnitude of the bed temperature so as to (i) avoid bed
agglomeration and incomplete char conversion (due to lower temperatures), (ii) avert tar formation (owing to high temperatures), and (iii) ensure high gasification efficiency. Conducting experiments, especially at a large scale, is often expensive, complicated and time-consuming task; modeling can save time and money.10 Owing to these advantages, a great deal of effort has been spent over the last five decades toward mathematically modeling different types of gasifiers. 1.1. Phenomenological Modeling of Fluidised Bed Coal Gasification. Commonly, phenomenological (“firstprinciples” or “mechanistic”) models of an FBCG are developed for gaining design and performance related information on the process operating under a variety of reaction conditions. These models incorporate complex and nonlinear reaction and mass and heat transport phenomena. Different types of models can be developed for the FBCG from the simple black-box or zero-dimensional models, where mass and heat balances are made over the entire gasifier to predict the exit gas composition, to the complex nonisothermal, three-dimensional ones taking into account the fluid dynamics and thermal behavior. Owing to the sheer complexity of the underlying physicochemical phenomena, the phenomenological modeling of the coal gasification/gasifier is a challenging task. This task is often simplified by making a number of assumptions regarding the numerous mechanisms underlying the gasification process. Broadly, two types of phenomenological models, namely thermodynamic (equilibrium) and kinetic (rate), are developed for the FBCG.11 The models belonging to the first category are independent of the gasifier type and assume complete oxygen consumption. Being independent of the gasifier type, these models are not useful for examining the effects of the operating parameters on the gasifier behavior. The other type, that is, kinetic models, comprises an appropriate hydrodynamic model of the fluidized bed coupled with the kinetics of the various reactions occurring in the gasifier. Given a set of operating conditions of a specific type of a gasifier, its kinetic model is capable of predicting the process behavior in terms of, for instance, product composition and temperature profiles. In the phenomenological modeling of an FBCG, once the model structure is fixed, then a large number of kinetic, thermodynamic, and heat and mass transport related parameters appearing in the model need to be determined either by conducting experiments or by simulation. Some notable representative studies as also reviews pertaining to the phenomenological modeling (including computational fluid dynamics modeling) of the fluidized bed gasification can be found in refs 12−28. The specific difficulties encountered in the phenomenological modeling of the gasification processes are (i) nonlinear interplay of multiple process variables, (ii) lengthy throughput dependent process dynamics, (iii) cost-intensive and exhaustive experimentation required for studying the effects of influential process operating variables and parameters on the process behavior, and (iv) unavailability of an in-depth knowledge of the physicochemical phenomena (e.g., kinetics, heat and mass transport mechanisms) underlying the coal gasification phenomena. 1.2. Alternate FBCG Modeling Strategies. An alternative approach to the phenomenological modeling of the gasification process is to utilize regression methods to formulate empirical models. However, in this approach the exact form of the datafitting function (model) needs to be specified before the 18679
dx.doi.org/10.1021/ie500593j | Ind. Eng. Chem. Res. 2014, 53, 18678−18689
Industrial & Engineering Chemistry Research
Article
function parameters could be estimated. This is a difficult task since in the gasification process multiple factors influence the nonlinear gasification phenomenon and the precise interactions between them are not fully known. The complexities involved in the phenomenological and regression-based modeling of the FBCG necessitate exploration of alternative nonlinear modeling strategies that do not require the full details of the physicochemical phenomena underlying the gasifier operation. The AI-based modeling strategies, for example, artificial neural networks (ANNs), and the statistical machine learning (ML) theory-based formalism, namely support vector regression (SVR), are exclusively data-driven strategies and thus can be used for modeling the FBCG. There exist a number of studies wherein ANNs have been employed in the energy-related science and engineering.29−33 In an exhaustive FBCG datadriven modeling study, Chavan et al.34 developed two ANNbased models for the prediction of gas production rate and heating value of the product gas, using the process data from the 18 globally located coal gasifiers. These models use six inputs namely, fixed carbon, volatile matter, mineral matter, air feed per kilogram of coal, steam feed per kilogram of coal, and temperature. Despite their potential, however, the AI and MLbased strategies have been only rarely employed in the modeling of fluidized bed gasifiers. Apart from the ANNs, the AI comprises a novel exclusively data-driven modeling formalism, namely genetic programming (GP). The uniqueness of the GP methodology is that given an example input-output data set, it is capable of searching and optimizing both, the specific form (structure) and the parameters, of an appropriate linear/nonlinear data-fitting functionand unlike ANNs, the GP does this without making any assumptions regarding the structure and parameters of the data-fitting function. Despite its novelty, the GP has not been used widely for the data-driven modeling applications in chemical engineering/technology to the same extent as the ANNs and SVR. Accordingly, the principal objectives of this paper are (i) to rigorously study the gasification of the high ash Indian coals in the pilot-plant scale FBCG and (ii) to develop GP-based models for the prediction of four FBCG performance variables, namely CO+H2 generation rate (y1) (kg/kg coal), syngas production rate (y2) (kg/kg coal), carbon conversion (y3) (%), and heating value of the syngas (y4) (kcal/Nm3). In the FBCG modeling, following eight process variables and parameters have been used as the model inputs: fuel ratio (fixed carbon/volatile matter) (x1), ash content of coal (x2) (wt %), specific surface area of coal (x3) (m2/g), activation energy of gasification (x4) (kJ/mol), coal feed rate (x5) (kg/h), gasifier bed temperature (x6) (°C), ash discharge rate (x7) (kg/h) and air/coal ratio (x8) (kg/kg coal). The novel features of the present study are as follows. (a) Extensive experimentation has been conducted for studying the high ash coal gasification under steadystate conditions in a pilot-plant scale FBCG located at the Central Institute of Mining and Fuel Research (CIMFR), Dhanbad, India. (b) A rigorous literature search shows that this is the first study wherein the GP strategy has been employed for the data-driven modeling in the coal-related energy science and engineering. (c) The principal component analysis (PCA) has been performed for reducing the dimensionality of the models’
eight-dimensional input space representing the various coal and gasifier parameters. (d) The sensitivity analysis of the eight model inputs has been performed to gauge their influence on the four process performance variables. Conventionally, phenomenological models for the coal gasification/gasifier use four types of inputs: (i) coal properties (proximate and/or ultimate analysis), (ii) process operating variables and parameters (reactor temperature, pressure, feed rate, etc.), (iii) physicochemical parameters (e.g., surface area) of the coal, and (iv) reaction kinetics parameters. In an earlier study on the FBCG modeling, Chavan et al.34 used the first two types of inputs. The present FBCG modeling study, in comparison, utilizes all the four types of inputs owing to which the input space now contains additional information pertaining to the physicochemical phenomena occurring in the gasifier. The importance of the selected eight model inputs (x1−x8) is described below. • The extents of fixed carbon and volatile matter reflect the effect of hydrogen and oxygen containing functional groups in the coal. This effect is represented in the form of fuel ratio (FC/VM) (x1) as a model input. • Ash (wt %) (x2) is an indicator of coal’s mineral matter content. Several formulas have been proposed for converting the ash content in the coal to its mineral matter content. One of the oldest yet still widely used correlation is given by Parr:35 mineral matter (wt%) = (1.08 × ash (wt%)) + (0.55 × S (wt%)) (1) where S refers to the sulfur content in the coal. Since the sulfur (as also carbonate) content of the Indian coals is low, a simplified version of the Parr correlation (eq 1) is used for these coals as given by (Choudhury36):
mineral matter (wt%) = 1.1 × ash (wt%)
•
•
• •
•
18680
(2)
The ash in the coal is responsible for lowering the carbonaceous material in the coal matrix thereby negatively affecting the quality of the product gas. The rate of gasification depends on the accessibility of the reactant gases to the internal surface of the porous coal where active sites reside. Accordingly, specific surface area (m2/g) (x3), has been considered as a model input. In the coal gasification process, the char−CO2 gasification is one of the rate controlling steps and hence the activation energy (kJ/mol) (x4) of the char−CO2 gasification reaction forms an input to the model. The coal feed rate (kg/h) (x5) has been chosen because it defines the flow rate of the basic carbonaceous raw material. The significance of the model input, namely gasifier bed temperature (°C) (x6), is that, as its magnitude increases, the product gas generation per kilogram of the coal increases. Also, higher gasification temperature results in the faster pyrolysis generating an increased amount of the CO2, which in turn gets converted to the CO via the Boudouard reaction. The ash discharge rate (kg/h) (x7) has been considered as a model input37 because together with the coal feed rate (x5), it significantly influences the residence time of dx.doi.org/10.1021/ie500593j | Ind. Eng. Chem. Res. 2014, 53, 18678−18689
Industrial & Engineering Chemistry Research
Article
Table 1. Analysis of Coal Samples (Air Dried Basis) proximate anal.
a
ultimate anal.
coal
ash (wt %) (x2)
moisture (wt %)
volatile matter (wt %)
fixed carbon (wt %)
C (%)
H (%)
N (%)
S (%)
O (%)a
C1 C2 C3 C4
41.3 48.9 27.0 36.0
6.5 7.1 9.7 8.1
24.5 20.4 25.7 20.7
27.7 23.6 37.6 35.2
37.15 30.82 48.46 43.51
2.83 1.90 3.44 3.03
0.86 0.60 1.03 0.98
0.55 0.24 0.60 0.51
6.68 5.55 7.07 4.27
By difference.
Figure 1. Fluidized bed gasification pilot plant consisting of process elements: (1) Coal feeding system. (2) Gasifying agent feeding system. (3) Fluidized bed gasifier. (4) Ash extraction system. (5) Cyclone separator. (6) Syngas cooling and cleaning system. (7) Flare stack.
comparison of their prediction and generalization performances, are provided. Finally, section 5 summarizes the principal findings of the study.
the coal particles in the gasifier bed. For instance, the residence time decreases with an increase in the coal feed rate (x5) or the ash withdrawal rate (x7). This results in the lower product gas generation per kilogram of the coal feed. Accordingly, in the low temperature gasification, it is necessary to allow a sufficient residence time for the coal particles to achieve maximum carbon conversion and product gas generation per kg of the coal (Chavan et al.,38 Kim et al.39). • The air/coal ratio (kg/kg of coal) (x8) is an important gasification process parameter since (Ocampo et al.;8 Kim et al.39): (a) the air-assisted oxidation of the carbon is one of the key reactions for attaining the desired temperature for the gasification, (b) an increase in the air/coal ratio increases the carbon conversion, and (c) an excessive air/coal ratio decreases the heating value of the syngas thereby negatively affecting the performance of the gasification process. In this study, the prediction accuracies and generalization capabilities of the four GP-based models have been compared with those of the corresponding four ANN-based models. This comparison indicates that both types of models possess an excellent ability to predict the magnitudes of the four gasifier performance variables. The structure of the present paper is as follows. The details of the FBCG and the experiments conducted thereof are given in section 2, followed by the description of the GP and ANN formalisms in section 3. In section 4, the results and discussion pertaining to (a) the sensitivity analysis of the eight model inputs, (b) the GP and MLP-based FBCG modeling, and (c) a
2. EXPERIMENTAL SECTION In this study, four types of Indian coals with ash content varying between 27% and 48% have been used. Their basic properties were evaluated by the proximate and ultimate analyses (see Table 1) carried out according to the Indian standards, viz. IS: 1350 (Part-I) 1984, IS: 1350 (Part-III) 1969, IS: 1350 (Part-IV/Sec-1) 1974, IS: 1350 (Part-IV/Sec-2) 1975. The specific surface area of the coal samples was measured using the Tristar 3000 surface area analyzer (Micromeritics, U.S.A.). Coal gasification comprises two steps: pyrolysis and char gasification.40 The kinetics of the char gasification was investigated in the reaction rate controlling regime. Here, char−CO2 gasification kinetic parameters, such as the gasification rate constant and the gasification activation energy thereof, were evaluated using the laboratory scale thermogravimetric analyzer (TGA) by following the procedure given in refs 41−43. 2.1. FBCG Pilot Plant. Gasification experiments were conducted in the air-blown FBCG pilot plant (Figure 1). The gasifier with a capacity to handle 10−20 kg coal/h and operating temperature
