Electricity load forecasting for smart home

Due to expanding global human population, the demand for electricity is ever increasing. Energy is difficult and expensive to store in bulk, so to ensure the demand can be met by energy suppliers, it is important to be able to forecast electricity load with high accuracy. Electricity load forecastin...

Full description

Saved in:
Bibliographic Details
Main Author: Lin, James Rizhong
Other Authors: Xu Yan
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2022
Subjects:
Online Access:https://hdl.handle.net/10356/163547
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:Due to expanding global human population, the demand for electricity is ever increasing. Energy is difficult and expensive to store in bulk, so to ensure the demand can be met by energy suppliers, it is important to be able to forecast electricity load with high accuracy. Electricity load forecasting (ELF) is a significant activity in topics such as power systems planning & operation. With the adoption of clean and sustainable energy, together with the need for more efficient and reliable power grids, there is a push towards Smart Grid. ELF is one of the relevant procedures made possible by Smart Grids. The aim of this project was to perform a long-term forecast of electricity load at the household level by using Prophet, a forecasting framework by Meta (previously Facebook). The dataset collected was of the average daily power consumption of a single residential home near France, Paris. Exogenous variables that affect energy usage, such as air temperature and humidity have also been added to the dataset. The results obtained were favourable, as it is apparent that the Prophet model had managed to capture the cyclic behaviour of the time series. The mean, median and seasonal Naïve models achieved a Mean Absolute Percentage Error (MAPE) of 33.493%, 32.995% and 29.492% respectively. The univariate Prophet model achieved an MAPE of 20.869%, and the univariate Prophet model with exogenous variables achieved an MAPE of 21.051%. After hyperparameter tuning of the latter, the final MAPE of 20.611% was achieved.