Development of a predictive model for enhanced grid densification and last mile connections in Uganda
Abstract
The United Nations sustainable development goal 7 ensures that by 2030, there is a universal
access to affordable, reliable and modern energy services. Electricity access is key for the social
and economic development of Uganda and the tracking of the load demand growth and electricity
infrastructure locations is key to making informed decisions on grid densification and last mile
connectivity. A case study of Nansana Municipality was chosen and the Namungoona Nansana
11kV feeder was analyzed. Data for the existing infrastructure, load profile data from Jan 2013 to
Feb 2023, and the reliability statistics data was obtained from UMEME for model development
and to understand the status and reliability of the existing network.
Load flow analysis was carried out on the simulated network using the Newton Raphson algorithm
and it converged with only three iterations indicating a good convergence behavior of the
algorithm. Reliability analysis was also performed to determine using statistical methods, the total
electric interruptions for loads on the feeder during an operating period and the reliability results
were within the standards set by the Electricity Regulatory Authority in the S-factor mechanism.
The loading on different line segments as well as the voltage levels on different busbars were also
analyzed and it was noted that the voltages at different buses especially those that were far away
from the substation and where the loads were concentrated were below the acceptable voltage
limits set by the authority (below ±10% of the nominal voltage). The voltage levels were improved
through network reconfiguration and conductor upgrade.
Two forecasting models were successfully developed and their performances were compared. An
LSTM model performed better than Random Forest model. The two models were combined to
obtain the resulting model known as the LSTM-RF model in order to improve the accuracy of the
predicted power and the LSTM-RF model outperformed both the LSTM and Random Forest
models across all the three evaluation metrics. Using the developed LSTM-RF model, the load
demand was predicated and it was observed that the maximum peak import power increased from
7.62MW to 8.46MW in a period of 3 years and this increased the loading on various line segments.
Also, the power losses increased from 0.94MW to 1.18MW at the end of the period.