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

net/publication/252047082

Coal Stock Pile simulation

Article · January 2011

CITATION READS

1 35

2 authors:

Vincent Micali Schalk Heunis

Stats4Buz (Pty) Ltd Environmental and Occupational Health Sciences Institute
6 PUBLICATIONS   2 CITATIONS    17 PUBLICATIONS   145 CITATIONS   

SEE PROFILE SEE PROFILE

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

Domestic Load Research View project

Electrical Energy Generation View project

All content following this page was uploaded by Schalk Heunis on 03 May 2016.

The user has requested enhancement of the downloaded file.

Coal stock pile simulation
by Dr. Vincent Micali and Dr. Schalk Heunis, Eskom

To safeguard the delivery of energy supply from coal fired power stations, (CFPS) a stock back up tonnage of coal is required at the station. This
stock is generally referred to as a stock pile (SP) whether measured in tonnage (SPT) or in days (SPD). The total stock reflects the stock of coal
(in tonnes) at hand, plus the replenishments (such as from stathes [e.g. bunker type] or direct-supply); therefore the total stock is equivalent to
the SPT. Essentially, the SPD reflects the daily SPD variation on the stock, based on the daily usage and replenishment.

The SPD’s were designed to fulfil most of

Eskom energy requirements for the coal
CFPSs, with a hedging position at acquiring
the shortfalls from the short to medium
term markets. However, with the growth in
ESO (Energy sent out), pressure has been
put on these markets to deliver the ever-
increasing shortfalls.
The lack of additional capacity (more
power stations) has also been a contributing
factor. Hence, the stock pile levels are at
risk and careful management of these has
become of outmost importance.

The Coal Stock Pile Simulator (CSPS) serves

as a simulation model which provides
Fig. 1: Example of results from the static model analysis.
“what-if ” scenarios and is forward looking.
It can provide scenario planning sets for
decision making. It is based on the ESO
forecasts (including monthly and time-
of-use profiles), the maintenance plan
(for the power stations), the modelled
forced outages (UCLF and OCLF), the
despatch plan (according to merit
order), the international imports (such as
Cahora Bassa) and the interrelationships
between Eskom’s nuclear, hydro, pump-
storage, OCGT and CFPSs. Since many
of the fundamental inputs are of a
stochastic (random) nature and Stochastic
Optimisation is still on the drawing board at
Universities all over the world, the CSPS is a
simulation model and not an optimisation Fig. 2: Dashboard methodology.
model but can be utilised as a heuristic
“optimiser ”.
The traditional way, was to set the minimum
level of SPD summarily at minimum of 20
days level while the maximum was set at
whatever the capacity of the stockyard
could take. However a statistical study
on the historical behaviours of the stock
piles indicated that each power station
exhibited different behaviours i.t.o. the
20 SPDs and the corresponding maxima.
The reason for the 20 SPD was that Fig. 3: Example of an output from the dynamic model run.
unexpected events could be addressed
within this time window and there was
enough capacity to cater for these events.
With the growth of ESO and demand,
a shrinking reser ve margin and not
enough capacity, the load fFactor at the
power stations has to increase which, by
inference, places pressure on the coal
contractual limitations, hence ancillar y
suppliers are required, which in turn come
with their risks such as availability, reliability
and costs, increasing the risks on the stock
pile. Fig. 4: Business interpretation of the dynamic model run.

energize - Jan/Feb 2012 - Page 17

GENERATION
Therefore the decisions required in
managing the stock piles need to come
from an informed position, not only from
a historical perspective, but also from a
forward looking one, supplemented with
“what-if ” scenarios.
These aspects are particularly important
for Eskom’s Primary Energy Division (PED)
as the latter is responsible for the planning,
procurement and delivery of coal to the
power stations.
Hence the instruments provided to PED,
need to enable the decision makers with
sufficient information, as close as possible
and feasible to reality.
In the early 2000s, a statistical model
called the static model was developed to
Fig. 5: Example of histogram of power station delivery.
deal with the different variation of the SPD’s
between different power stations. The static
model works on the historical behaviour
of the stock pile with the latest behaviour
having more weight than the old one. It
is modelled on a power station basis and
alarm levels are based on the variations of
its estimates. The model may be applied
at a station level to provide station targets,
with the assumption that the past repeats
itself. As an example, Fig. 1 shows the
variation of the stock pile days per station
in historical graph form and the actual
(bar) level for a particular financial year
end. Some of the stations show a below
the ‘Lo limit’ condition which indicates an
alarm situation. The methodology used
was of a statistical nature [2, 3].
The static model expresses the coal
Fig. 6: Box and whisker plot of delivery reliability
(Actual/Budget). stockpile system level as an aggregate of
all stations. This is not very meaningful as an
indicator of how well the system as a whole
is doing, for example, one station may
have a stockpile level of 10 days and the
other 80 days – resulting in an aggregate
of 45 days. These stock pile estimates can
be aggregated at system level only for an
indicative purpose.

Dealing at system level, due the factors

mentioned above, a more appropriate
technique would be to use a “Dashboard”
methodology. This is a different type of
aggregating the per formance of the
Fig. 7: Implementation process. different stock piles at their corresponding
power stations. This is illustrated in Fig. 2.
Using this method, the dashboard is used
to calculate a weighted score where poor
performance is weighted more to ensure
that problem areas are not absorbed
within the aggregate.

A dynamic model (can be used in the

same fashion as the static model, with the
advantage that “what-if” scenarios can be
considered. This removes the constraint
that histor y has to repeat itself, and
provides the planner with a tool to evaluate
the plan and adjust as necessary. It should
be noted that the dynamic model is based
on simulation and not optimisation, i.e. it
can evaluate a plan, but neither produce
nor ensure an optimum plan. The Coal
Fig. 8: Results of sensitivities. Stock Pile Simulator (CSPS) is essentially a

energize - Jan/Feb 2012 - Page 18

 GENERATION
dynamic model based on Monte Carlo simulations. During
the simulation, thousands of iterations of possible stockpile
level futures are generated, based on the specified
probabilistic model. Various statistics can be compiled
based on these futures, for example the 5th and 95th
percentile and expected values as shown in Fig. 3.
The upper and lower bounds can be used to set targets
to monitor the actual performance of the stockpile levels.
For example, the 5th percentile stockpile (red line in Fig. 3)
can be utilised to signal an alarm level condition.
The actual stockpile level should ideally stay in the buffer
and overshoot zones; these are the normal variations
one may expect as less coal is delivered or more coal is
burnt. The actual should only move into the base due to
extreme events and could result in having catastrophic
consequences. Movements into the strategic region could
occur due to hedging or other supply specific strategies.
It should be noted that the previous two figures were
compiled with data available in March 2007 for the power
stations and the production plan of the same month.

Analytical methodology
The dynamic model
In order to construct a simulation model that provides
realistic future scenarios, it is necessary to understand
the causes and effects in the power generation system.
This section documents some of the cause and effect
diagrams that were developed, the data that was
collected and their treatments.
The dynamic model consist of a consolidation of various
modules.
Here the modelling of various elements of transporter
reliability is illustrated (from “Transporter behaviour” to “Skills
shortages”).
Other similar modules are coal sources, coal quality, power
stations parameterisations.
Data collection and statistical techniques

Data was sourced from two main source systems (the

Primary Energy and the Generation Divisions systems) and
then inserted into the data processing module. Exact
dumps of tables were taken, except where the source
of the data was spreadsheets. In this instance data
was imported directly into the data processing module
and then manipulated in order to get it into the desired
format. The data validation process forms part of the data
processing module. Data needs to conform to rules that
are compliant to business processes. The latter forms the
basis to formulate the rules for validation. Fig. 7 gives an
illustration of the business process.

The coal stockpile simulator contains a number of

statistical models which represent the uncertainty inherent
in the production in coal plans, for example the delivery
of coal. Changes in the stockpile level are related to
these uncertainties, through a set of equations based on
operational processes a set of statistical models is given
below:
Stock pile level = Coal delivered – coal burnt
Coal delivered = Coal contract ± delivery
uncertainty
Coal burnt = Energy generated x heat rate/CV
CV = Expected CV ± CV uncertainty
Energy generated = Energy scheduled – load losses +
Additional load
Additional load = Load forecast uncertainty +
operational reschedule

energize - Jan/Feb 2012 - Page 19

GENERATION
As an example, the behaviour of “Coal Various validation tests proved to be iteratively refined as understanding about
delivered” (see above) at a power station adequate and the simulator models were the subtleties involved is obtained. This
is illustrated in Fig. 6 which shows a accepted. Also “ impact assessments” permits in-depth understanding of the
histogram of the difference between the was performed in order to determine if behavioural patterns of the coal stock pile.
actual and planned delivery for the various certain changes that would be required,
stations. The p-value of the Shapiro-Wilk Statistical models were derived to represent
as well as their corresponding downstream
[2] test shows that in most cases, for the the uncertainty in various components of
effects of such changes, may have
power stations the null hypothesis that the system. These models are first order
on the overall implementation of the
these distributions are normal, cannot be models and could be refined in future
simulation. Thereafter a testing process
rejected at 95% confidence level. phases of this initiative. A sensitivity analysis
during which, changes were screened
was also performed on the existing model.
Fig. 6 by using Box & Whisker methodology to determine whether they were feasible
[6], illustrates the behaviour of the station or not and whether or when they could The model was implemented in an
deliver y reliability. The box shows the be implemented considering project open source decision support framework
interquartile spread of the error away timelines. The agreed changes were then (DIAS) using Microsoft Excel files as the
from expectation (i.e. the Budget). The implemented as per project-plan. After the primary configuration and data exchange
dotted-line extending from the box, shows changes implementations, the simulator mechanism. The application was iteratively
the extension of the error to the 5 and 95 was then tested extensively before being developed to ensure rapid deliver y of
percentiles; while the dots (2 in Fig. 6 for presented to the intended user base, for features and accommodating changes
the "road") show the extreme values. piloting (2011/12). in focus. The CSPS has demonstrated
This reliability measure would therefore its value at various stages of piloting
Training and transfer of skills formed an
show the change, based on the mixture of especially in contributing to the plans of
integral part of the project-plan. The
transport types. The reliability is modelled, the new generation of coal fired plant in
statistician (PED) in charge of the CSPS,
in this case, as the percentage error South Africa.
was fully trained on the assumptions,
and expressed as: actual/budget. The derivations, sources, models, modules and Acknowledgments
quantification of the estimates for the operational aspects of the CSPS.
station reliability parameters could therefore The authors are grateful for the inputs and
be updated to reflect the transport method. It is envisaged that the CSPS will be in full support of the Primary Energy Division of
operation by the end of 2012 or early Eskom with particular votes of thanks to
Models sensitivities play an important role 2013. Dan Marokane, Steven MacDougall, Andre’
in simulation. A way to measure these is
van Heerden, Athisha Hardeyal. Also our
by using a first order multiple regression Results
gratitude goes to Dr. Chris van Alphen for
model [7]. This model was utilised to assess
The methodology used for simulation was his contribution toward the quality aspects
the sensitivities on the modules "Inputs"
to the module "Output". For instance, the method of Monte Carlo simulation and of the coal for the corresponding module.
for the transporter reliability (output), the the sensitivity one was multiple regression We are also indebted to Chris Schutte
inputs would be the range from done for (first order). general manager technology (previously
the overall coal stock pile simulator in acting divisional executive PED) who was
The sensitivities’ results indicate that, for
assessing which inputs in various modules the initial sponsor of this project. We would
the percentage change in mean stock
have the largest effect on the mean coal like to thank Barry MacColl (acting general
pile, the major contributing inputs are: the
stock pile days, percentage change manager RT&D) for backing this research
"Heat-rate", the "CV", the "Net dDelivery",
(SPD%). till its implementation.
the "Energy sent out" and the "Maintenance
The model used was of the form [7]: outages PCLF" (in that order). This paper was presented at the 2011
Industrial and Commercial use of Energy
Conclusion Conference, Cape Town August 2011 and
where, the xij represent the inputs’s values The dynamic model, based on Monte is reprinted with permission.
for i iterations and j=i,…,n input variables
Carlo simulation, allows planners to
References
Implementation per form “ what-if ” analysis to evaluate
[1] V Micali, “Energy Flow Simulator ” Eskom
different plans and scenarios.
The Generation production plan is a crucial Research Activity PRJ10-00051100-3668,
The outputs can also be used as control Energy Markets: Economics, Statistical and
input into the Primary Energy coal plan. It
Decision Sciences Centre of Excellence, Energy
determines, given the ESO forecasts, the l i m i t s, i. e. a l a r m l e v e l s t o c o m p a r e Economics, Statistical & Decision Sciences
dispatch required from each of the power actual performance against the planned Steering Committee 2010/11.
stations. It declares the coal required at performance. [2] J W Tukey, “Exploratory data Analysis” Reading
given heat-rates. This in turn forms the input Mass: Addison-Wesley, 1977.
The dynamic model is not an optimisation
to the coal plan which incorporates the [3] F Mosteller, J W Tukey, “Data Analysis and
model but provides probable upper and Regression” Addison-Wesleyt, 1977
coal quality data. This information is then
lower bounds to the performance of each [4] S S Shapiro, M B Wilk, (1965). “An Analysis
stored in the Eskom data warehouse (EDW).
station. At the heart of the simulation of Variance Test for Normality (Complete
This process is illustrated in Fig. 7.
model is a cause-and-effect diagram Samples)”. Biometrika 52 (3-4): 591–611.
Once the inputs are drawn from the that was constructed to understand the [5] V Micali, S Heunis, “Stockpile Simulation”, Eskom
EDW, the coal stockpile input file can be inter relationship between the different Research Report, April 2011
amended by the user and it may contain components of the system and was used [6] W W Hines, D C Montgomery, “Probability and
additional information, for example the Statistics in Engineering and Management
as a road map for iterative improvements Science”, Wiley, 3rd Ed., 1990
parameters of the statistical models. The to the model. A subset of the interactions [7] J Neter, W Wasserman, “Applied Linear Statistical
amended input file is then loaded into the has been implemented to date and this Models”, Richard D. Irwin Inc., 1974.
stockpile simulator. After the simulation process will continue into future phases of
is completed, a report with the results Contact Dr. Vincent Micali, Eskom,
the project.
is generated, analysed and used for Tel 072 783-7721,
decision support. The cause-and-effect diagram will be [email protected] 

energize - Jan/Feb 2012 - Page 20

View publication stats