Abstract:
Accurate forecasting is becoming increasingly important due to the energy
consumption's rapid rise. To manage and plan the use of energy resources efficiently,
it is crucial to predict the demand for electricity, whose high increase in Uasin Gishu
City has had a negative impact on the reliability of the electrical supply. The utility
company KPLC) presently forecasts medium-term electricity demand through feeder
load checks to support decision-making on operations, maintenance or infrastructure
development planning, but this has not been sufficient to address the impact on power
stability. Therefore, the main objective of this study was to model and simulate a hybrid
model to forecast short-term electricity demand in Uasin Gishu County. The specific
objectives were to determine the short-term electricity demand profile as affected by
weather variables, time effects, economic factors and Load parameters, apply an Hybrid
model based on Adaptive Neuro-fuzzy inference System (ANFIS) to estimate load
demands from an hour to a week ahead, and assess the system’s performance. Using
temperature, wind speed, humidity, and historical load data as the primary parameters,
this study describes the development and application of an ANFIS-based STLF model
for the power networks in the Uasin Gishu County. Past load data from Kenya's
electricity networks and meteorological data from the cloud data base
www.timeanddate.com were used to test and validate the model. An adaptable neuro-
fuzzy inference system (ANFIS) was used for machine learning-based electricity
predictions. The forecasting model was constructed using a total of 49,860 dataset
points, with training accounting for 75% of the work and checking and validation
accounting for 15%. The novelty of this research lies in the large quantity of availed
data, input parameters, validation of the ANFIS model for the training, testing, and
validation data using four different membership functions: triangular, trapezoidal,
generalized bell shaped, and Gaussian curve shaped, which produced the mean absolute
percentage error (MAPE) values of 0.588, 0.359, 0.671, and 0.567, respectively. The
effectiveness of the suggested approach is demonstrated by the evaluation of trained
FIS results and a separate set of data based on Uasin Gishu county's electricity demand
estimates. The suggested model's efficacy is clearly demonstrated by its average mean
absolute percentage error of 0.0997%. The acquired results and forecasting
performance demonstrate the viability of the suggested strategy and demonstrate the
significant influence of meteorological variables on the short-term load demand profile.