| dc.description.abstract | The main aim of an electric power system is to reliably provide electrical power supply to
customers. However, in 2020, Uganda’s grid experienced an average of 62.4% forced
outages. The forced outages due to generator failures led to a loss of 20.36GWh. Such
failures during peak time can cascade into blackouts affecting plant availability. Therefore,
it is important to estimate the availability of units and forecast their production to ensure
continuity of power supply to the grid.
As such, the objective of this research was to develop computational methods for
availability estimation with a case study of Nalubaale Power Station (NPS). First, the
generation and outage trends for power generating units at NPS were analysed. NPS was
observed to run as a base load power plant with an average and maximum dispatch of
96.66MW and 145.3MW respectively. Then, a model was developed for availability
estimation of NPS units using the Bootstrap Monte Carlo Simulation (BMCS) method. The
model provided acceptable results as an approximate method with a Mean Square Error
(MSE) as low as 0.001 for Unit 2 and the highest MSE of 0.092 for Unit 9. Based on the
results, a merit order for dispatch of the units was formed; First Unit 1, then 4, 8, 5, 7, 2 and
9 as the last unit to be synchronized.
Lastly, a model for forecasting hydro power electricity generation was developed using
Long-Short Term Memory (LSTM) Machine Learning (ML) technique. The model achieved
an MSE of 0.0067. The model makes an hour ahead prediction based on 24-hour historical
data for the turbine discharge and head. From the results of this study, Eskom Uganda
Limited (EUL) and other concessionaries running other hydro power stations should adopt
the BMCS model developed for availability estimation of their generating units. Results
from this model will aid in deciding whether to retain or acquire a new generator system and
in determining a merit order of units. Furthermore, Uganda Electricity Transmission
Company Limited (UETCL) should change the requirement in Power Purchase Agreements
(PPAs) for hydro power stations to declare their available capacity mainly based on power
generation forecast instead of expected planned outages. This process could utilize the
LSTM model developed in the study. Overall, the proposed approach provides a more
accurate availability estimate that can aid in power system planning and maintenance. | en_US |
