Electric Energy Demand-Supply Optimization by Utilizing Ensemble Learning Framework

Electricity demand prediction is a crucial aspect of calculating energy consumption,

especially for meeting the required amount of energy between supply and demand

is indispensable for energy manufacturers. However, current prediction models do

not effectively consider the impact of weather, leading to low prediction accuracy

and large deviations in Bangladesh. Consequently, the energy demand is associated

with various challenges such as climate, adjusting generation distribution and so

on. Hence, electric industries have paid attention to short-term energy forecasting

to assist their management system. To address these challenges, multiple machine

learning and ensemble models are compared in this paper. Firstly, six ML models

named Random Forest Regressor (RFR), Support Vector Regressor (SVR), Extreme

Gradient Boosting Regressor (XGBoost), K-Nearsest Neighbor (KNN), Ridge Re-

gression and ElasticNet are compared. Then two ensemble models are compared,

one used voting ensemble regressor and the other used stacking ensemble regressor.

This data-driven approach uses Bangladeshi electricity consumption data collected

by a power distribution company in Dhaka and daily maximum and minimum tem-

perature data collected from BMD. Our proposed ensemble model outperforms all

baseline models and the voting ensemble regression model giving the lowest MAPE

of 4.86%. It shows statistically significant performance and indicates this approach

is promising for forecasting short term electricity consumption.

• Supervision