Download as pdf or txt
Download as pdf or txt
You are on page 1of 37

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/223512503

Calculation models for prediction of Liquefied Natural Gas (LNG) ageing


during ship transportation

Article  in  Applied Energy · May 2010


DOI: 10.1016/j.apenergy.2009.10.023

CITATIONS READS

74 6,338

5 authors, including:

Mario Miana Rafael del-Hoyo-Alonso


ITAINNOVA Instituto Tecnológico de Aragón
20 PUBLICATIONS   393 CITATIONS    49 PUBLICATIONS   251 CITATIONS   

SEE PROFILE SEE PROFILE

Vega Rodrigálvarez José R. Valdés


Instituto Tecnológico de Aragón Instituto Tecnológico de Aragón
8 PUBLICATIONS   121 CITATIONS    21 PUBLICATIONS   362 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

PolicyCLOUD - Policy management through technologies across the complete data lifecycle on cloud environments (H2020) View project

Grapevine. High performance computing services for prevention and control of pests in fruit crops. View project

All content following this page was uploaded by Mario Miana on 22 December 2017.

The user has requested enhancement of the downloaded file.


Calculation models for prediction of Liquefied Natural Gas
(LNG) ageing during ship transportation.

Mario Miana, a,* Rafael del Hoyo, a Vega Rodrigálvarez, a José Ramón Valdés, a
Raúl Llorens, b
a
Instituto Tecnológico de Aragón, Área de Investigación, Desarrollo y Servicios Tecnológicos,
María de Luna 7, 50018 Zaragoza, Spain.
b
ENAGAS S.A., Dirección de Ingeniería y Tecnología del Gas, Autovía A – 2, km. 306.4, 50012
Zaragoza, Spain.
* Corresponding author. Tel: +34 976 01 11 57; fax: +34 976 01 18 89; E-mail address:
mmiana@ita.es (M. Miana).

Abstract.

A group of European gas transportation companies within the European Gas Research

Group launched in 2007 the ‘MOLAS’ Project to provide a software program for the analysis of

the liquefied natural gas (LNG) ageing process during ship transportation. This program contains

two different modeling approaches: a physical algorithm and an ‘intelligent’ model. Both models

are fed with the same input data, which is composed of the ship characteristics (BOR and

capacity), voyage duration, LNG composition, temperature, pressure, and volume occupied by

liquid phase at the port of origin, together with pressure at the port of destination. The results

obtained are the LNG composition, temperature and liquid volume at the port of destination.

Furthermore, the physical model obtains the evolution over time of such variables en route as it

is based on unsteady mass balances over the system, while the i – model applies neural

networks to obtain regression coefficients from historical data composed only of origin and

destination measurements. This paper describes both models and validates them from previous

published models and experimental data measured in ENAGAS LNG regasification plants.

Keywords: liquefied natural gas ageing, physical model, neural networks, LNG ship carriers.

Nomenclature.

e error, defined in equation 26, dimensionless;

E energy (kJ);

1
F vapour fraction of a LNG mixture in equilibrium, dimensionless;

H enthalpy (kJ/kmol);

m iteration;

P pressure (mbar);

Q quality;

Q heat flow (kW);

t time (h);

T temperature (ºC);
3
V volume (m );
3
VTk ship capacity (m );

x weighted sum of the input synapses, dimensionless;

X molar fraction of a component of the liquid phase of LNG mixture, dimensionless;

y output of the ith neuron, dimensionless;

Y molar fraction of a component of the vapour phase of LNG mixture, dimensionless;

Z molar fraction of a component of LNG mixture, dimensionless.

Greek symbols.

 convergence tolerance, defined in equation 6;

 selected variable;

 phase;

 density (kg/m );
3

 difference.

Subscripts.

cal calculated from physical algorithm or i – model;

d day;

dst destination;

eq equilibrium;

ev evaporated;

i component;

2
imod calculated from i – model;

ini initial;

L liquid;

meas measured in LNG regasification plants;

ori origin;

out gas flow taken out of the tank to avoid large increments of pressure;

phys calculated from physical model;

V vapour.

Acronyms.

BOR Boil – Off Rate, defined in equation 1, dimensionless;

C1 methane;

C2 ethane;

C3 propane;

ETLV function to calculate equilibrium temperature of a LNG liquid – vapour mixture;

FL function to calculate the vapour fraction of a LNG liquid – vapour mixture;

GERG European Gas Research Group;


3
HHV Higher Heating Value (kWh/m );

iC4 iso – butane;

iC5 iso – pentane;


3
LD liquid density (kg/m );

LNG Liquefied Natural Gas;

LDF function to calculate liquid density;

LH function to calculate liquid enthalpy;

MLP Multilayer Perceptron;

MOLAS MOdels for prediction of LNG Ageing during Ship transportation;

N2 nitrogen;

nC4 n – butane;

nC5 n – pentane;

VDF function to calculate vapour density;

3
VH function to calculate vapour enthalpy;

Wobbex Index (kWh/m ), WI  HHV


3
WI LD .

1. Introduction.

Liquefied Natural Gas (LNG) is an energy source with a worldwide steady growth: it is

expected to almost double in size between 2005 and 2010, delivering around 40% of global gas

supply growth. This increment of LNG demand has provoked a fast augmentation of LNG

transport by sea. At the end of 2006, 208 LNG ships were in operation and it is planned that, at

the beginning of 2010, the LNG fleet will number a minimum of 326 ships [1].

LNG is a liquid mixture of light hydrocarbons, with methane (C1) as the main component at,

approximately, atmospheric pressure. LNG is also composed of other minority components such

as nitrogen (N2), ethane (C2) and propane (C3), and traces of i – butane (iC4), n – butane

(nC4), i – pentane (iC5) and n – pentane (nC5).

LNG is transported by ship at atmospheric pressure close to its boiling point which is

around –162 ºC (111 K) depending on LNG composition. In spite of the high degree of insulation

in the tank walls, it is impossible to avoid a net input heat transfer from the surroundings so that

vaporization will be always present during LNG transportation by ships. Vaporization provokes

an increase in pressure in the tank, so a certain amount of the vapour phase is taken out of the

tank to avoid dangerous overpressure. Usually, this outlet gas flow is used by the ship itself to

reduce its fuel consumption. Figure 1 describes the LNG tank dynamics during ship

transportation.

Owing to the different boiling points of LNG components (varying from -196 ºC to +36 ºC),

LNG vaporization is not homogenous: components with the lowest boiling point (N2 and C1)

tend to evaporate in a more important way than heavy components. This phenomenon, which

does not occur in pipeline transport of natural gas, is called ageing or weathering and its main

consequence is the modification over time of LNG composition and properties. The accurate

4
prediction of LNG composition before unloading in regasification plants has two main benefits:

the prevention of possible accidents caused by stratification and roll over phenomena in ground

tanks, and the necessary support for carrying out blending or mixing operations to meet quality

specifications for transport, distribution and utilization of natural gas.

To the authors’ knowledge, the unsteady evolution of LNG composition during ship

transportation is a rarely studied subject covered by a very reduced number of references.

Kountz [2] measures the evolution over time of LNG composition, temperature and mass

contained in a pressurized tank under a constant heat flow transferred to the mixture and

proposes a physical model to predict this phenomena. Aspelund et al [3] have developed a

physical models to predict the LNG ageing phenomena under constant heat flow in small LNG –

chains. Both models are based on equilibrium between liquid and vapour phases, yielding a

good approximation of experimental data measured in [2]. Other interesting references are the

vaporization studies by Boe [4] and Conrado and Vesovic [5] concerning LNG on water and the

investigation of the rollover phenomenon in storage tanks by Bates and Morrison [6].

The scarcity in scientific literature about LNG weathering contrasts with its importance for

the above-mentioned issues and the growing number of publications about a broad range of

LNG subjects such as facilities ([7] – [9]), safety ([10] – [14]), atmospheric dispersion ([15], [16]),

modeling of spills and Rapid Phase Transitions (RPT) from water or land surfaces ([17] – [27])

and thermal topics associated with fire scenarios ([28] – [35]). A common approach in these

references is to consider LNG as a substance defined by certain equations of state or, simply, by

C1 properties, neglecting the influence of changes in chemical composition.

In this context, a group of European gas companies led by ENAGAS started the GERG

Project 1.59 ‘MOLAS’ with the aim of building a software program for predicting changes in LNG

composition during ship transportation from loading port (origin) to unloading port (destination).

Current MOLAS program can anticipate LNG composition and properties such as Higher Heat

Value (HHV), Wobbe Index (WI) and Liquid Density (LD) at the port of destination using the

information available in certificates on loading from LNG ships or general measurements that

can easily be obtained during ship unloading at LNG regasification plants.

5
The MOLAS program contains two different modeling options:

- a physical model based on mass balances and equilibrium state between liquid and

vapour phases;

- an ‘intelligent’ model (i – model), based on artificial neural networks to take into

account nonlinearities deriving from the specific characteristics of each ship, quality,

or port of origin or destination.

The selection of one or other of the models depends on several factors, principally the LNG

quality (defined as the origin country or port where the LNG is loaded) and the available

historical data stored in a specific database developed to handle the principal information

relating to LNG trading by ship.

This paper consists of five sections. The physical model is described in section 2, while

section 3 explains the i – model. Both models are applied in section 4 to several test cases, and

their accuracy compared with respect to data measured in ENAGAS regasification plants.

Finally, section 5 summarizes the conclusions of the study.

2. The physical model.

The physical model is based on mass balances to predict LNG composition, temperature

and volume at the port of destination from the initial conditions, ship characteristics and voyage

duration. The following sections summarize the main hypothesis and the different blocks that

compose the algorithm.

2.1. General hypothesis underlying the physical model.

The following hypotheses are assumed:

1. The algorithm is independent of the shape and number of tanks in the ship. Two

parameters define the ship characteristics: the Boil – Off Rate (BOR) and the ship

capacity, defined as the total LNG volume that can be carried by the ship.

6
2. The LNG mixture is in equilibrium state between liquid and vapour phases at boiling

(equilibrium) temperature during the voyage. Subcooled or superheated states in

tanks are not considered. In fact, it is assumed that the initial conditions are

transformed into equilibrium conditions at the beginning of the first time step.

3. Constant time step size: the evolution over time of the ageing process is discretely

calculated in time steps of 0.5 h size.

4. BOR is defined as the quotient of the vaporised volume per day to LNG volume at

port of origin:

Vev ,d
BOR  (1)
VL,ori

A typical value is 0.15 %, although the exact magnitude depends on the type of

container and the quality of its thermal insulation.

It is assumed that changes in LNG density are small during ship transportation

because composition and temperature remain nearly constant. Thus BOR can be

arranged in a molar basis according to equation 2:

nev ,d
BOR  (2)
nL,ori

Evaporated moles from equation 2 are obtained, by definition, per day. If it is

supposed that they are constant during the voyage, equation 3 calculates the

number of evaporated moles in one time step:

nev ,d
nev ,t  (3)
48

5. Linear evolution of pressure over time: the evolution over time of tank pressure is

assumed to follow a linear relationship from origin to destination conditions:

Pdst  Pori
P  Pori  t  tori  (4)
t dst  t ori

2.2. Main blocks of the algorithm.

7
The algorithm of the physical model is composed of four parts: the reading block, the initial

block, the unsteady block and the writing block, as shown in figure 2.

The reading block is executed first in order to obtain the input data that defines the LNG

ageing process to be simulated: ship characteristics (capacity and BOR), trip duration (tdst), initial

filling percentage (%L,ori), pressure (Pori), temperature (Tori) and composition of liquid phase at

port of origin (Xori) together with pressure at port of destination (Pdst).

Next, the origin liquid composition, temperature, pressure, ship capacity and filling

percentage are used in the initial block to calculate the initial equilibrium conditions at the loading

port, as described in section 2.2.1. After that, these initial equilibrium conditions, the BOR and

the destination pressure are sent to the unsteady block described in section 2.2.2 in order to

calculate the evolution over time of composition, temperature and filling percentage by means of

an iterative loop. At the end of the execution of the MOLAS application, two different sets of

results are stored by the writing block: the composition, pressure, temperature and liquid volume

at port of destination and the evolution over time of these variables during the voyage.

LNG equilibrium constants and temperature, vapour fraction and liquid and vapour densities

or enthalpies are calculated through the functions described in table 1, based on the Lee – Erbar

– Edminster state equations ([36], [37]), the solution method of Rijkers and Heidemann [38] and

the model of McCarty [39].

2.2.1. Initial block.

The loading process in LNG liquefaction plants consists of filling ship tanks near to their

capacity, typically around 98 %. Operating conditions (pressure and temperature) are near to

LNG boiling point at atmospheric pressure. It is assumed that the measured composition at the

loading port is the composition of the liquid phase at temperature Tori and that the tank is only

filled with liquid. At the end of the loading, the liquid evaporates to fill the free volume of the ship.

The initial block calculates the equilibrium temperature at which the LNG load from the port

of origin, composed only of liquid phase whose composition is Xi,ori, tends to occupy the defined

8
percentage of the volume of the tank at the operating pressure. This calculation is performed by

the iterative method shown in figure 3.

In the first iteration (m = 0), the equilibrium temperature of the LNG mixture being sought is

estimated from the temperature at the port of origin with an increment initialized to 0.02 ºC. This

estimated temperature together with the mixture loading composition and the operating pressure

are used to estimate the vapour fraction (Fm) and the constants of equilibrium between liquid and

vapour phases (Ki,m) by means of the FL function.

The composition of liquid (Xi,m) and vapour (Yi,m) phases at equilibrium temperature Tm are

then obtained from the XY function and these compositions are applied in LDF and VDF

functions to obtain the liquid (L,m) and vapour (V,m) densities, respectively. Next, the percentage

of the volume occupied by liquid (%L,m) is calculated according to equation 5:

V ,m 1  Fm 
%L,m 
V ,m 1  Fm    L,m Fm
(5)

This percentage is used to define the convergence criterion of the iterative method shown
–4
in equation 6, where the convergence tolerance  is set to 10 :

%L,ori  %L,m   (6)

If the convergence criterion is satisfied, the estimated equilibrium temperature is directly set

as the initial equilibrium temperature. If the criterion is not satisfied, the iterative loop is repeated,

at a maximum of 200 times, relating the new guess estimate of the temperature increment with

the convergence criterion according to equation 7:

T  0.02%L,ori  %L,m  (7)

At the end of the execution of the initial block, the algorithm has calculated the LNG

equilibrium temperature, the composition of the LNG mixture, the compositions of the liquid and

the vapour LNG phases, and the liquid and vapour moles that fully occupy the ship capacity.

9
2.2.2. Unsteady block.

The unsteady block is the main part of the algorithm. It starts with the output data from the

initial block and obtains, at the end of each time step, the LNG composition, equilibrium

temperature and liquid and vapour volumes by means of an iterative method. These output data

are fed back again to repeat calculations for the next time step, until the voyage duration is

completed.

For any time step, the unsteady block starts with the evaluation of the operating pressure

from the current time by equation 4. Next, liquid and vapour moles at the beginning of the time

step (nL,t and nV,t) are calculated from the ship capacity (VTk), the percentage occupied by liquid

phase (%L,t) and the liquid and vapour densities, which are estimated from the LDF and VDF

functions. After that, BOR is applied to obtain the evaporated moles considering equation 3.

Now it is necessary to compute liquid, vapour and mixture compositions and equilibrium

temperature at the end of the current time step. This is done by an iterative loop based on mass

balances 8 and 9 for liquid and vapour phases, respectively:

nL,t  t  nL,t  nev ,t (8)

nV ,t  t  nV ,t  nev ,t  nout ,t (9)

nout,t being the vapour moles taken out of the tank during the current time step to avoid large

increments of pressure. Then, liquid and vapour moles at the end of the current time step are

defined from density and volume occupied by each phase:

n L,t  t  V L,t  t (10)

nV ,t  t  V V ,t  t (11)

The sum of liquid and vapour volumes is always VTk, so equations 8 to 11 can be

rearranged into the mass balance for moles taken out of the tank as shown in equation 12:

V ,t  t    BOR
nout ,t  nV ,t  V ,t  tVTk  nL,t  1  V ,t  t  n (12)
L,t  t  L,t  t  48 L,ori
 

10
The solution of equation 12 requires liquid and vapour densities at the end of the current

time step, which can be calculated by LDF and VDF functions if the equilibrium temperature and

the liquid and vapour compositions are known at this time. Thus, the iterative loop shown in

figure 4 is applied.

The first iteration (m = 1) assumes that the equilibrium temperature at the end of the time

step is 0.01 % higher than the temperature at the beginning of the time step and there are no

changes in the LNG mixture composition:

Tm,t  t  Tt  0.0001 Tt (13)

Z i ,t  t  Z i ,t (14)

These hypotheses allow the vapour fraction (Ft+t) and the equilibrium constants (Ki,t+t) to

be estimated by means of the FL function at the end of the time step. With this information,

liquid and vapour densities can be estimated, so a first approach to equation 12 can be

provided. Once equation 12 is solved, the mass balances of equations 8 and 9 yield the liquid

and vapour moles present at the end of the time step. These data are next applied to a mass

balance of liquid phase to obtain equation 15 as an estimated approach for liquid composition:

nL,t  nV ,t K i ,t
X i*,t  t  X i ,t
K i ,t  t nV ,t  t  nOut ,t   nL,t  t
(15)

where estimated variables are denoted by superscript (*).

The estimated liquid composition is now used to obtain vapour and mixture compositions

from the previously calculated vapour fraction and equilibrium constants, according to equations

16 and 17:

Yi *,t  t  X i*,t  t K i ,t  t (16)

Z i*,t  t  1  Ft  t X i*,t  t  Ft  t Yi *,t  t (17)

*
The ETLV function computes an estimated equilibrium temperature (T m,t+t) from the
*
estimated mixture composition (Z i,t+t) and operating pressure (Pt+t). The estimated equilibrium

temperature is then fed back to the beginning of the iterative process, as shown in equation 18:

11
Tm1,t  t  Tm* ,t  t (18)

This iterative loop converges fast, so it is repeated 5 times with no convergence criterion.

Finally, liquid and vapour volumes are actualised at the end of the current time step from

equations 19 and 20:

n Ev ,t
VL,t  t  VL,t  (19)
 L,t

VV ,t  t  VTk  VL,t  t (20)

The state at the end of the current time step is now completely defined by liquid, vapour and

mixture compositions, operating pressure, equilibrium temperature and liquid and vapour

volumes. All this information, together with LNG properties (HHV, WI and LD), is stored for

further analysis of the evolution over time of the ageing process and fed back to repeat the

unsteady block.

2.3. Comparison with experimental data and published models.

The evolution over time of composition and temperature predicted by MOLAS physical

model is compared for validation purposes with the experimental results measured in [2] and the

predicted values by physical models included in [2] and [3].

This validation exercise simulates the LNG weathering in a pressurized tank of 190 liters of

capacity with a catalytic heater to accelerate the LNG ageing. Six weathering tests are

performed in [2], varying the input parameters as the inlet heat flow, the LNG composition, the

initial load or the test duration. Tests no. 5 and 6 are selected for comparison with MOLAS

application. Table 2 contains the input data that define both test cases, together with the

estimation of BOR, which is calculated from the variation of the mass of the tank at the initial and

intermediate points. The obtained values are 4.0 and 3.1 for tests 5 and 6, respectively,

representing a vigorous ageing process compared to the typical BOR of 0.15 during ship

transportation.

12
Figures 5 and 6 compare the evolution over time of composition, temperature and pressure

for both tests, yielding a good agreement among the three physical models and experimental

results, especially for test 6. However, the results of MOLAS physical model for test 5 show a

lower vaporization and a larger content in C1. Although the three physical models are based on

equilibrium between liquid and vapour phases, the models included in [2] and [3] calculate the

unsteady evolution from a constant heat flow added to the tank to solve an energy balance, but

the LNG evaporation in MOLAS physical model is simplified by assuming a constant evaporation

rate along the ageing process, approximate by the BOR parameter.

This approximation can be evaluated by the heat flow transferred to the tank during each

time step, which can be calculated by equation 21 in the MOLAS model:

Et  t  Et  Eout
Q  (21)
t

being Et and Et+t the energy stored inside of the tank at the beginning and at the end of the

time step, respectively, and Eout the energy leaving the tank by the vented stream to avoid large

increments of pressure. These energy terms are estimated from the liquid, vapour and vented

mols and liquid and vapour enthalpies, calculated by LH and VH functions, according to

equations 22 and 23:

Et  t  n L ,t  t H L ,t  t  nV ,t  t H V ,t  t (22)

E out,t  nout,t H V ,t (23)

The average heat flows obtained from MOLAS physical model are 11.54 W and 10.20 W

for tests no. 5 and 6, respectively, which are 21 % and 13 % lower than the input heat flows

applied in [2] and [3]. This difference can be provoked by the intermediate point to estimate the

BOR: for test nº 5, this point is located in the middle of the period, but test nº 6 provides the

mass of the tank at an intermediate point located near to the end of the weathering process.

Therefore, the hypothesis of constant evaporation rate assumed by the use of the BOR

parameter should be improved in future developments. However, it can be considered an

13
acceptable approach for ageing processes if BOR is calculated from data at the end of the

ageing process.

3. The intelligent model (i – model).

3.1. Introduction.

Learning from examples (or historical data) is one of the abilities that makes artificial neural

networks a suitable approach for the modeling of LNG ageing. Supervised neural models, such

as the well – known Multilayer Perceptron (MLP) [40], have demonstrated their capability for

solving supervised regression problems in many applications. In chemical and electrochemical

knowledge areas, several problems have been solved using neural networks, such as sensor

discrimination between butane and propane for gas detection [41], prediction of the strains in

gas generators in a liquid rocket engine [42], identification of a chemical process reactor control

for nonlinear dynamical systems [43], or the static modeling of an electrochemical PEM

(Polymer Electrolyte Membrane) fuel cell [44]. A common application of neural networks is in

forecasting. They are often used for stock prediction and have also been used in gas

consumption prediction [45].

It is important to remark that traditional approaches, such as linear or polynomial regressive

methods, can provide reasonable results for regression models. However, in real situations, the

information required for modeling LNG ageing is frequently either not available in its entirety or is

not acquired at the same time. Incompleteness, a high degree of heterogeneity of information

used, or high dimensionality of input information are common problems with real data.

Therefore, taking into account all these considerations for modeling LNG ageing, a neural

network is considered a better approach than traditional statistical techniques.

In the MOLAS application, an MLP neural network approach is used to model the LNG

ageing process by calculating from historical data the temperature, volume and molar fraction of

LNG components at the port of destination. This model is called the i – model (intelligent model)

and is composed of two blocks, the Generation block and the Execution block, as shown in

figure 7.

14
3.2. The Generation block.

The Generation block calculates the regression model from the historical data of the LNG

ageing process. These historical data are stored in a database included in the MOLAS

application. This database is editable with tools available in the MOLAS program itself to

increase the number of registers for storing data of future LNG voyages.

The regression model is based on 10 neural networks: eight of them estimate the molar

fraction of the LNG components (C1, C2, C3, iC4, nC4, iC5, nC5 and N2) at the port of

destination, while the remaining two are devoted to obtaining the temperature and liquid volume

at the port of destination from the input data. These input data are the LNG composition,

temperature, pressure and liquid volume at the port of origin, trip duration, BOR and quality, and

pressure at the port of destination. Additional input variables have been studied, for example the

ship’s name, but these have not been included because the lack of sufficient data for these input

variables lengthens the time required to obtain the i – model with no significant improvements in

the results.

Figure 8 shows a schematic description of the neural network composed of 6 neurons in

one layer for evaluation of the molar fraction of C1. The number of neurons for each network is

determined by ‘trial and error’, increasing the number of neurons until no significant

improvement is achieved.

Input data are connected to these neurons, which are modeled by non linear activation

functions shown in equations 24 and 25:

y  tanhx  (24)


y  1 e  x 
1
(25)

Here y is the output of the ith node (neuron) and x is the weighted sum of the input

synapses. In this program, the neural network is trained by the Backpropagation Training

Algorithm, which adjusts the weightings of the connections in response to the output errors of

the network. During training, the output values of the network being trained are compared to the

desired output values from the training set, and an error is estimated. This error is then

15
propagated backwards (the backpropagation part of the algorithm) over the network and used to

calculate changes to the connection weights.

The estimated error is calculated using a 10 fold cross validation technique. This method

can be defined as the statistical practice of partitioning a sample of data into subsets such that

the analysis is initially performed on a single subset, while the other subset(s) are retained for

subsequent use in confirming and validating the initial analysis. The initial subset of data is

called the training set; the other subset(s) are called validation or testing sets.

In the MOLAS application, the full sample of historical data about LNG ageing is partitioned

into 10 subsamples. A single subsample from these 10 subsamples is retained as the validation

data for testing the model, and the remaining 10 − 1 subsamples are used as training data. The

cross–validation process is then repeated 10 times (the folds), with each of the 10 subsamples

used exactly once as the validation data. The 10 results from the folds are then averaged to

produce a single estimation of the error obtained from the i – model.

This error is compared to the error obtained from the physical model, that is, the differences

between the estimation of the physical model and the values measured at the port of destination.

In fact, both errors from the physical algorithm and from the i – model are obtained from

equation 26:

meas  cal
e  100 (26)
meas

where meas is the measured value stored in the historical database, while cal is the value

predicted by the physical algorithm or by the i – model.

If the error of the i – model is less than the error of the physical model for the prediction of

the molar fraction of a determined LNG component (N2, C1, C2…), temperature or liquid

volume, the i – model is selected for the prediction of the variable in question. In contrast, if the

error of the i – model is larger than the error of the physical model, the physical model is

selected for forecasting the quantity.

16
In sum, the use of the i – model implies the execution of the physical model also, because

the results obtained with the i – model are a selection of the best modeling options from the

information of the estimated errors for both approaches over the historical data. Finally, all the

neural trained models together with the estimated errors are saved in the database contained in

the MOLAS program. This neural network model will be used in the Execution block.

3.3. The Execution block.

The Generation block obviously requires a lengthy execution time due to the high number of

mathematical operations to be calculated. However, it is not necessary to calculate again the i –

model (and the errors obtained) every time the i – model is executed. In fact, the neural network

model must be regenerated only when new data are included in the database. Therefore, the

execution block directly applies the previously generated i – model in three steps. First, the i –

model is recovered from the database. Next, the solution for the input data is calculated from

selected physical and i - models for each variable. Finally, the results for each factor are

obtained from the best estimated model. This procedure avoids having to generate the neural

network every time the i – model is used.

4. Ageing prediction.

This section compares the ageing prediction obtained from the physical algorithm and the i–

model with selected registers obtained from historical cargo measurements at ENAGAS

regasification plants located in Spain. According to table 3, 96 % of unloaded LNG in Spain is of

Qualities 1 to 5, so one example is selected from each quality. Tables 4 to 8 compare the results

obtained with both models.

For space reasons, figure 9 shows only the evolution over time for the results (composition,

temperature, filling percentage Wobbe Index and energy content of the ship) obtained with the

physical model applied to the example of Quality – 1. To ensure a legible reading of the change

of each variable, MOLAS results are properly represented in unitary form dividing current value

by the value at port of origin. It is observed that the molar fractions of C1 and N2 experience a

constant decrease, specially significant for N2 being reduced to the 65 % of its initial content.

17
Although the molar fraction of C1 keeps nearly constant (the decrease is 0.1 %), the energy

content of the ship (E), calculated as the number of liquid mols multiplied by the HHV, suffers a

decrement of nearly 2 % due to the poorer LNG quality and the fall of the volume occupied by

the liquid phase.

It must be noted that, from the initial measured data, there is no nC5 at the loading port.

However, there is a low concentration of nC5 in the measured composition at the port of

destination. This difference probably results from the varying accuracy of measurement systems

at the ports of origin and destination and therefore can not be predicted by the physical model.

However, the i – model can handle such non linearities and it predicts a certain amount of nC5

at the port of destination.

Next, the accuracy of both models is tested by analysing the differences between their

results and experimental values of selected variables measured during 153 LNG cargos in

ENAGAS regasification plants during 2008. These data were not included when building the i –

model, so the reliability of this modeling option is tested for new cargos. The number of cargos

analysed for each quality and the average trip duration (measured in days) are shown in figure

10.

The accuracy study is performed over the following selected variables: molar fraction of C1,

HHV, WI and LD. The relative error, also defined by equation 26, is limited for each variable to

the maximum values compiled in table 9 [46, 47]. These limits represent the maximum allowable

differences in measurement procedures for LNG composition and properties. Hence, if the

differences between measured values in regasification plants and the predicted results by the

MOLAS program are lower than these limits, the MOLAS estimation is within the admitted

uncertainty of the measurement itself.

Figure 11 shows the average of the errors obtained for each quality. Predictions of C1 and

WI meet the imposed limits in all qualities, but HHV and WI tend to be underpredicted by both

models, especially for Quality 1. In any case, the errors obtained are of the same order of

magnitude as the limiting criteria so it can be concluded that both models are accurate enough

to predict the LNG ageing process during ship transportation.

18
Both models are compared in table 10 which shows the percentage of accurate predictions

for each variable, that is the percentage of predictions from the physical algorithm and the i –

model that meet the limiting criteria. The i – model gives better results than the physical model

except for some results for Q3 and Q4, which are both characterized by short durations. In fact,

the extremely short duration of trips from Q3 minimizes the differences between both models.

5. Conclusions

Two different modeling approaches have been developed to predict LNG ageing during

ship transportation: a physical model and an ‘intelligent’ model. Errors obtained from both

options are in the range of the accuracy limits proposed by [46] and [47].

The physical model is based on mass balances and thermodynamic equilibrium between

liquid and vapour phases during the voyage, while the i – model predicts the destination

composition, temperature and volume by regression of historical data with neural networks. In

general, a better performance by the i – model is observed. Thus, it is expected that the

accuracy of the i – model will be increased in future years as the number of available cargos to

build it will be augmented by the growth of the LNG market. Besides, it is probable that the

increasing database will change the configuration of the neural network with new variables or

different numbers of neurons or layers.

In any case, the physical model is always a valid option if the i – model can not be applied

due to the lack of historical data for new ships, routes or qualities.

Results provided by the MOLAS program can be used in three main areas of the LNG

industry:

 Operative area: the knowledge, in advance, of LNG density to be unloaded may

help terminal operators to predict stratification into ground tanks and consequently

reduce future roll over incidents.

 Quality area: sometimes, a LNG property such as WI does not meet the severe

quality requirements imposed in some countries for the transport, distribution and

19
utilization of natural gas. The MOLAS application helps quality personnel to decide

necessary actions (blending & mixing) to fulfill quality specifications.

 Financial area: although the MOLAS program is not intended for billing purposes, it

may also be used to estimate the amount of energy to be transferred from ship to

ground tanks.

To summarise, the MOLAS program is a user friendly, powerful and reliable tool that allows

the prediction of the composition of LNG to be unloaded at its destination, in other words the

most essential LNG properties as WI, HHV and energy content. The MOLAS program can help

terminal operators to manage regasification plants in a safer and more efficient manner.

Acknowledgements.

The authors greatly appreciate the comments, suggestions and support received from

Angel Benito, Concepción Rabinal and Luis Carlos Gutiérrez (ENAGÁS, Dirección de Ingeniería

y Tecnología del Gas), Luisa Shelenko and Paul Martin (Advantica), Abdallah Touil and Noela

Vazquez (Gaz de France – Suez, Research and Development Division), Oddvar Jørstad

(StatoilHydro, Oil and Gas Refining, R & T), Peter Schley (E.On Ruhrgas, Thermodynamics /

Gas Quality Group), Ángel M. Gutiérrez (Naturgas Energía) and Robert Coll (REPSOL, LNG &

GTL Technology).

References.

[1] International Energy Agency. Natural Gas Market Review. Paris; 2006.

[2] Kountz KJ. Weathering of on – board storage tanks. Project final report, Institute of Gas

Technology, IGT Project 32034 – 02, 1999.

[3] Aspelund A, Gjøvåg GA, Nekså P, Kolsaker K. LNG – chain, a calculation tool for natural

gas quality in small scale LNG distribution chains, CR06 – 133, ICEC – 21, Prague, 2007.

[4] Boe R. Poll boiling of hydrocarbon mixtures on water, International Journal of Heat and

Mass Transfer 1998; 41 : 1003 – 1001.

20
[5] Conrado C, Vesovic V. The influence of chemical composition on vaporization of LNG and

LPG on unconfined water surfaces, Chemical Engineering Science 2000; 55 : 4549 – 4562.

[6] Bates S, Morrison DS. Modeling the behaviour of stratified liquid natural gas in storage

tanks: a study of the rollover phenomenon, International Journal of Heat and Mass Transfer

1997; 40 : 1875 – 1884.

[7] Chen QS, Wegrzyn J, Prasad V. Analysis of temperature and pressure changes in liquefied

natural gas (LNG) cryogenic tanks, Cryogenics 2004; 44 : 701 – 709.

[8] Havens J, Spicer T. United States regulations for sitting LNG terminal: problems and

potential, Journal of Hazardous Materials 2007; 140 : 439 – 443.

[9] Taylor DW. The role of consequence modeling in LNG facility sitting, Journal of Hazardous

Materials 2007; 142 : 776 – 785.

[10] Koopman RP, Ermak DL. Lessons learned from LNG safety research, Journal of

Hazardous Materials 2007; 140 : 412 – 428.

[11] Cleaver P, Johnson M, Ho B. A summary of some experimental data on LNG safety,

Journal of Hazardous Materials 2007; 140 : 429 – 438.

[12] Moore DA, Fuller B, Hazzan M, Jones JW. Development of a security vulnerability

assessment process for the RAMCAP chemical sector, Journal of Hazardous Materials

2007; 142 : 689 – 694.

[13] Lee DH, Kim MH, Kwon SH, Kim JW, Lee YB A parametric sensitivity study on LNG tank

sloshing loads by numerical simulations, Ocean Engineering 2007; 34 : 3 – 9.

[14] Vanem E, Antao P, Ostvik I, del Castillo de Comas F. Analysing the risk of LNG carrier

operations, Reliability Engineering & System Safety 2008; 93 : 1328 – 1344.

[15] Pereira JCF, Chen XQ. Numerical calculations of unsteady heavy gas dispersion, Journal

of Hazardous Materials 1996; 46 : 253 – 272.

[16] Luketa – Hanlin A, Koopman RP, Ermak DL. On the application of computational fluid

dynamics codes for liquefied natural gas dispersion, Journal of Hazardous Materials 2007;

140 : 504 – 517.

21
[17] Thyer AM. A review of data on spreading and vaporization of cryogenics liquid spills,

Journal of Hazardous Materials 2003; A99 : 31 – 40.

[18] Fay JA. Models of spills and fires from LNG and oil tankers, Journal of Hazardous Materials

2003; B96 : 171 – 188.

[19] Qiao Y, West HH, Manna MS, Johnson DW, Cornwell JB. Assessment of the effects of

release variables on the consequences of LNG spillage onto water using FERC models,

Journal of Hazardous Materials 2006; 130 : 155 – 162.

[20] Luketa – Hanlin A. A review of large – scale LNG spills: Experiments and modeling, Journal

of Hazardous Materials 2006; A132 : 119 – 140.

[21] Hissong DW. Keys to modeling LNG spills on water, Journal of Hazardous Materials 2007;

140 : 465 – 477.

[22] Woodward JL. Coupling dynamic blow down and poll evaporation model for LNG, Journal

of Hazardous Materials 2007; 140 : 478 – 487.

[23] Spaulding ML, Swanson JC, Jayko K, Whittier N. An LNG Release, transport and fate

model system for marine spills, Journal of Hazardous Materials 2007; 140 : 488 – 503.

[24] Vesovic V. The influence of ice formation on vaporization of LNG on water surfaces,

Journal of Hazardous Materials 2007; 140 : 518 – 526.

[25] Johnson DW, Cornwell JB. Modeling the release, spreading and burning of LNG, LPG and

gasoline on water, Journal of Hazardous Materials 2007; 140 : 535 – 540.

[26] Fay JA. Spread of large LNG polls on the sea, Journal of Hazardous Materials 2007, 140 :

541 – 551.

[27] Gavelli F, Bullister E, Kytomaa H. Application of CFD (FLUENT) to LNG spills into

geometrically complex environments, Journal of Hazardous Materials 2008; 159 : 158 –

168.

[28] Lehr W, Simecek – Beatty D. Comparison of hypothetical LNG and fuel oil fires on water,

Journal of Hazardous Materials 2004; 107 : 3 – 9.

22
[29] Fay JA: Model of large pool fires, Journal of Hazardous Materials 2006; B136 : 219 – 232.

[30] Raj PK. Large hydrocarbon fuel pool fires: Physical characteristics and thermal emission

variations with height, Journal of Hazardous Materials 2007; 140 : 280 – 292.

[31] Raj PK. LNG fires: A review of experimental results, models and hazard prediction

challenges, Journal of Hazardous Materials 2007; 140 : 444 – 464.

[32] Pitblado R. Potential for BLEVE associated with marine LNG vessel fires, Journal of

Hazardous Materials 2007; 140 : 527 – 534.

[33] Raj PK. LNG pool fire spectral data and calculation of emissive power, Journal of

Hazardous Materials 2007, 142 : 720 – 729.

[34] Havens J, Venart J. Fire performance of LNG carriers insulated with polystyrene foam,

Journal of Hazardous Materials 2008, 158 : 273 – 279.

[35] Suardin JA, Wang Y, Willson M, Mannan MS. Field experiments on High Expansion (HEX)

Foam application for controlling LNG pool fire, Journal of Hazardous Materials, In Press,

Corrected Proof, (2008), doi:10.1016/j.hazmat.2008.10.040.

[36] Lee BL, Erbar JH, Edmister WC. Liquefied Natural Gas: Thermodynamic properties at low

temperatures, Chemical Engineering Progress 1972; 68 : 83 – 84, 99.

[37] Lee BL, Erbar JH, Edmister WC. Prediction of thermodynamic properties for low

temperature hydrocarbon process calculations, AIChE Journal 1973; 19 : 349 – 353.

[38] Rijkers MPW, Heidemann RA. Convergence behaviour of single – stage flash calculations,

ACS Symposium Series 1986; 300 : 476 – 493.

[39] McCarty RD. Four Mathematical Models for the Prediction of LNG Densities, NBS

Technical Note 1030, Boulder (USA), 1980.

nd
[40] Haykin S. Neural Networks: a Comprehensive Foundation. 2 ed. Prentice-Hall, Englewood

Cliffs, New Jersey (USA), 1999.

23
[41] Morsi I. Discrimination between butane and propane in a gas mixture using semiconductor

gas sensors and neural networks, Proc. IEEE Sensors Applications Symposium SAS 2008,

2008, 134–139.

[42] Li F, Deng C, Song S, Duan J. Prediction of the Strains in Gas Generators Based on BP

Neural Networks, Proc. Pacific-Asia Workshop on Computational Intelligence and Industrial

Application PACIIA '08, 2008, 23–26.

[43] Al-Hiary H, Braik M, Sheta A, Ayesh A: Identification of a chemical process reactor using

soft computing techniques, Proc. (IEEE World Congress on Computational Intelligence).

IEEE International Conference on Fuzzy Systems FUZZ-IEEE 2008, 2008, 845–853.

[44] Hatti M, Tioursi M, Nouibat W. Static Modelling by Neural Networks of a PEM Fuel Cell,
nd
Proc. IECON 2006 - 32 Annual Conference on IEEE Industrial Electronics, 2006, 2121–

2126.

[45] Peharda D, Delimar M, Loncaric S. Short term hourly forecasting of gas consumption using
rd
neural networks, Proc. 23 International Conference on Information Technology Interfaces

ITI 2001, 2001, 367–371.

[46] ENAGAS: Especificación técnica EV 203 de sistemas analíticos para análisis en continuo y

discontinuo de gas natural para facturación, Revisión 4, Madrid (Spain), 2009.

nd
[47] GIIGNL: Custody Transfer Handbook, 2 ed. International Group of Liquefied Natural Gas

Importers, Levallois (France), 2001.

24
FIGURE CAPTIONS
a)

b)

Figure 1. LNG tank dynamics: a) initial state at time t b) final state at time t + t.

25
Figure 2. The physical model.

Figure 3. Iterative algorithm of the initial block.

26
Figure 4. Iterative algorithm of the unsteady block.

27
a)

b)

Figure 5. Comparison of MOLAS physical model with experiments ([2]) and published models
([2] and [3]) for test 5: a) Composition b) Temperature and pressure.

28
a)

b)

Figure 6. Comparison of MOLAS physical model with experiments ([2]) and published models
([2] and [3]) for test 6: a) Composition b) Temperature and pressure.

29
Figure 7. The i - model.

Figure 8. Simplification diagram of the architecture used to build the statistical model: neural
network for estimation of C1 at destination port.

30
a) b)

c)

Figure 9. Evolution over time of results of example from Quality – 1: a) molar fraction of C1, C2,
C3, iC4, nC4 and iC5 in liquid phase b) molar fraction of N2 in liquid phase c) temperature,
percentage of filling, WI and energy value.

Figure 10. Number of analysed cargos and average trip duration.

31
a)

b)

Figure 11. Average errors: a) Physical model b) i – model.

32
TABLE CAPTIONS

Table 1. Thermodynamic functions.

Function name Input data Output data Comment


Provides the equilibrium temperature for a liquid –
ETLV Zi , P Teq
vapour LNG mixture at a given pressure.
Provides the equilibrium constants, the vapour fraction
of a LNG mixture and the phase (liquid, vapour or
FL Zi, P, Teq F, Ki, 
mixture) for a given composition, operating pressure
and equilibrium temperature.
Provides liquid density of a LNG liquid phase for a given
LDF Xi, P, Teq L
operating pressure and equilibrium temperature.
Provides liquid enthalpy of a LNG liquid phase for a
LH Xi, P, Teq HL
given operating pressure and equilibrium temperature.
Provides vapour density of a LNG vapour phase for a
VDF Yi, P, Teq V
given operating pressure and equilibrium temperature.
Provides vapour enthalpy of a LNG vapour phase for a
VH Yi, P, Teq HV
given operating pressure and equilibrium temperature.
Provides the composition of liquid and vapour phases
XY Zi, F, Ki Xi, Yi from the composition, the vapour fraction and the
equilibrium constants of a LNG mixture.

Table 2. Input data for tests 5 and 6.

Test no.
Variable 5 6
C1 91.9 % 87.8 %
C2 6.8 % 6.8 %
C3 1.3 % 1.0 %
Initial gas composition iC4 0.0 % 0.0 %
(Molar fraction) nC4 0.0 % 0.0 %
iC5 0.0 % 0.0 %
nC5 0.0 % 0.0 %
N2 0.0 % 4.4 %
Tank pressure 7700 mbar 7700 mbar
Heat flow 14.64 W 11.71 W
Initial loaded LNG mass 59 kg 66 kg
Initial filling percentage 81.42 % 91.44 %
LNG mass at intermediate point 29.5 kg 13.6 kg
Time at intermediate point 13 days 25 days
Time duration of ageing experiment 21 days 27 days
BOR 4.0 3.1

Table 3. Percentage of incoming qualities in ENAGAS regasification plants.

Quality Percentage
1 34.4 %
2 18.4 %
3 17.9 %
4 16.7 %
5 8.6 %
6 3.1 %
7 0.5 %
8 0.3 %

33
Table 4. Ageing prediction for a cargo example from Quality – 1.

Trip duration 258 h


3
Capacity 134487 m
BOR 0.15

Pressure
Origin 1090 mbar
Destination 1185 mbar

Variable ori, meas dst, meas dst, phys edst, phys dst, imod edst, imod
N2 0.065 % 0.035 % 0.043 % -18.60 % 0.047 % -25.53 %
C1 92.753 % 92.605 % 92.658 % -0.06 % 92.612 % -0.01 %
C2 4.843 % 4.789 % 4.922 % -2.70 % 4.945 % -3.15 %
C3 1.976 % 2.037 % 2.008 % 1.44 % 2.002 % 1.75 %
iC4 0.191 % 0.260 % 0.194 % 34.02 % 0.220 % 18.18 %
nC4 0.161 % 0.265 % 0.164 % 61.59 % 0.161 % 64.60 %
iC5 0.011 % 0.008 % 0.011 % -27.27 % 0.009 % -11.11 %
nC5 0% 0.001 % 0% -- 0.004 % -75.00 %
Temperature -159.5 ºC -158.7 ºC -158.8 ºC -0.06 % -158.8 ºC -0.06 %
3 3 3 3
Volume 129969 m 127168 m 127894 m -0.57 % 127513 m -0.27 %
%L 96.6 % 94.6 % 95.1 % -0.57 % 94.814 % -0.27 %
3
WI (kWh/m ) 15.336 15.367 15.347 0.13 % 15.350 0.11 %
3
HHV (kWh/m ) 11.914 11.914 11.928 -0.12 % 11.936 0.23 %
3
LD (kg/m ) 446.697 446.832 446.042 0.18 % 446.224 0.14 %

Table 5. Ageing prediction for a cargo example from Quality – 2.

Trip duration 390 h


3
Capacity 137661 m
BOR 0.15

Pressure
Origin 1084 mbar
Destination 1140 mbar

Variable ori, meas dst, meas dst, phys edst, phys dst, imod edst, imod
N2 0.360 % 0.186 % 0.193 % -3.13% 0.146 % 27.40 %
C1 90.300 % 90.142 % 90.273 % -0.15% 90.100 % 0.05 %
C2 6.160 % 6.399 % 6.315 % 1.33% 6.495 % -1.48 %
C3 2.250 % 2.300 % 2.289 % 0.48% 2.291 % 0.39 %
iC4 0.370 % 0.389 % 0.370 % 5.14% 0.403 % -3.47 %
nC4 0.550 % 0.578 % 0.550 % 5.09% 0.550 % 5.09 %
iC5 0.010 % 0.005 % 0.010 % -50.00% 0.012 % -58.33 %
nC5 0% 0.001 % 0% -- 0.003 % -66.67 %
Temperature -159.9 ºC -159.9 ºC -159.7 ºC 0.13% -158.8 ºC 0.69 %
3 3 3 3
Volume 136089 m 133147 m 132816 m 0.25% 132553 m 0.45 %
%L 98.9 % 96.7 % 96.5% 0.25% 96.3 % 0.45 %
3
WI (kWh/m ) 15.439 15.494 15.483 0.07% 15.506 -0.08 %
3
HHV (kWh/m ) 12.178 12.236 12.217 0.18% 12.247 -0.09 %
3
LD (kg/m ) 457.035 457.71 456.891 0.18% 456.236 0.32 %

34
Table 6. Ageing prediction for a cargo example from Quality – 3.

Trip duration 98 h
3
Capacity 30000 m
BOR 0.15

Pressure
Origin 1100 mbar
Destination 1125 mbar

Variable ori, meas dst, meas dst, phys edst, phys dst, imod edst, imod
N2 0.715 % 0.383 % 0.615 % -37.72 % 0.463 % -17.28 %
C1 87.417 % 87.722 % 87.450 % 0.31 % 87.539 % 0.21 %
C2 8.950 % 9.018 % 9.007 % 0.12 % 9.079 % -0.67 %
C3 2.226 % 2.210 % 2.236 % -1.16 % 2.238 % -1.25 %
iC4 0.286 % 0.276 % 0.286 % -3.50 % 0.275 % 0.36 %
nC4 0.370 % 0.355 % 0.370 % -4.05 % 0.370 % -4.05 %
iC5 0.019 % 0.020 % 0.019 % 5.26 % 0.019 % 5.26 %
nC5 0.017 % 0.016 % 0.017 % -5.88 % 0.016 % 0.00 %
Temperature -161.4 ºC -160.1 ºC -160.9 ºC -0.50 % -159.8 ºC 0.19 %
3 3 3 3
Volume 28818 m 28748 m 28643 m 0.37 % 28672 m -0.27 %
%L 96.1 % 95.8 % 95.5 % 0.37 % 95.6 % -0.27 %
3
WI (kWh/m ) 15.467 15.528 15.489 0.25% 15.519 0.06%
3
HHV (kWh/m ) 12.311 12.347 15.050 0.37% 15.086 0.13%
3
LD (kg/m ) 465.735 462.959 464.973 -0.43% 463.188 -0.05%

Table 7. Ageing prediction for a cargo example from Quality – 4.

Trip duration 126.5 h


3
Capacity 138500 m
BOR 0.15

Pressure
Origin 1128 mbar
Destination 1138 mbar

Variable ori, meas dst, meas dst, phys edst, phys dst, imod edst, imod
N2 0.030 % 0.028 % 0.030 % -6.67% 0.039 % -28.21 %
C1 97.180 % 97.294 % 97.156 % 0.14% 97.581 % -0.29 %
C2 2.480 % 2.410 % 2.504 % -3.75% 2.102 % 14.65 %
C3 0.170 % 0.156 % 0.170 % -8.24% 0.170 % -8.24 %
iC4 0.060 % 0.057 % 0.060 % -5.00% 0.052 % 9.62 %
nC4 0.030 % 0.029 % 0.030 % -3.33% 0.030 % -3.33 %
iC5 0% 0.019 % 0.000 % -- 0.020 % -5.00 %
nC5 0.050 % 0.007 % 0.050 % -86.00% 0.008 % -12.50 %
Temperature -159.8 ºC -159.8 ºC -160.0 ºC -0.120% -159.6 ºC 0.13 %
3 3 3 3
Volume 136102 m 134984 m 135034 m -0.04% 134486 m 0.37 %
%L 98.3 % 97.5 % 97.5 % -0.04% 97.1 % 0.37 %
3
WI (kWh/m ) 15.039 15.031 15.042 -0.07% 15.014 0.11 %
3
HHV (kWh/m ) 11.367 11.347 11.367 -0.18 % 11.322 0.22 %
3
LD (kg/m ) 429.596 429.052 429.964 -0.21% 427.982 0.25 %

35
Table 8. Ageing prediction for a cargo example from Quality – 5.

Trip duration 283.5 h


3
Capacity 140500 m
BOR 0.17

Pressure
Origin 1117 mbar
Destination 1126 mbar

Variable ori, meas dst, meas dst, phys edst, phys dst, imod edst, imod
N2 0.011 % 0.020 % 0.011 % 81.82% 0.029 % -31.03 %
C1 96.691 % 96.519 % 96.635 % -0.12% 96.427 % 0.10 %
C2 2.758 % 2.931 % 2.814 % 4.16% 2.973 % -1.41 %
C3 0.447 % 0.448 % 0.447 % 0.22% 0.447 % 0.22 %
iC4 0.042 % 0.032 % 0.042 % -23.81% 0.038 % -15.79 %
nC4 0.033 % 0.036 % 0.033 % 9.09% 0.033 % 9.09 %
iC5 0.018 % 0.008 % 0.018 % -55.56% 0.008 % 0.00 %
nC5 0% 0.006 % 0% -- 0.005 % 20.00 %
Temperature -159.6 ºC -159.8 ºC -159.9 ºC -0.06% -159.6 ºC 0.13 %
3 3 3 3
Volume 137936 m 135144 m 135199 m -0.04% 134121 m 0.76 %
%L 98.2 % 96.2 % 96.2 % -0.04% 95.5 % 0.76 %
3
WI (kWh/m ) 15.075 15.078 15.075 0.02% 15.078 0.00 %
3
HHV (kWh/m ) 11.422 11.433 11.428 0.04% 11.436 -0.03 %
3
LD (kg/m ) 431.079 431.809 431.685 0.03% 431.726 0.02 %

Table 9. Maximum limits to errors.

Variable Limit Reference


C1 0.5 % [46]
HHV 0.25 % [46]
WI 0.25 % [46]
LD 0.27 % [47]

Table 10. Percentage of accurate predictions.

Q1 Q2 Q3 Q4 Q5
Physical i– Physical i– Physical i– Physical i– Physical i–
model model model model model model model model model model
C1 74 % 87 % 83 % 95 % 100 % 100 % 100 % 100 % 88 % 94 %
HHV 7% 24 % 43 % 62 % 65 % 70 % 95 % 63 % 76 % 82 %
WI 67 % 76 % 93 % 95 % 78 % 74 % 100 % 100 % 88 % 100 %
LD 28 % 50 % 50 % 81 % 91 % 91 % 88 % 100 % 76 % 94 %

36

View publication stats

You might also like