Abstract:
Manufacturing industry is a key contributor to the economy.
These manufacturing firms are high energy consumers and currently facing
challenges in their energy demand due to scarcity, cost, subsequent
environmental impact on heavy consumption, social and consumer pressure
for carbon footprint of consumer goods and industry specific global
compliances and regulations. Then an accurate short-term forecast of
energy consumption is a must to maintain optimal supply and usage while
minimizing negative concerns against higher consumption. This will enable
a smooth operation with a minimum risk on energy related inventories.
Electricity, biomass, furnace oil and diesel are the main energy sources in
industry. In current context, studies are conducting on electricity
consumption in industries, buildings, and residential boundaries. However,
predictions on total energy consumption on manufacturing firms are not
frequently studied. Both conventional statistical models and deep learning
models are widely used for this task. In this exercise, a manufacturing entity
has been selected to predict its forthcoming month’s energy consumption
using historical energy consumption and manufacturing figures. Existing
literature related to energy consumption prediction suggested, four models
to predict energy consumptions. Fb prophet, vector auto regression, Long
Sort Term Memory (LSTM) and Auto Regressive Integrated Moving Average
(ARIMA) models were taken to make predictions for energy consumption in
terms of electricity and biomass. Performance of each models were
compared using root mean squared error, mean absolute error and mean
percentage absolute error. LSTM outperformed over all other models for both
types of energies and the same model was selected to predict monthly
energy consumption of selected entity.
