Load forecasting using machine learning.
Abstract
Short term load forecasting is essential within low income developing countries for power
system planning, operation and control. In this project, load forecasting is done using a non
linear Regression Analysis through implementation of artificial neural networks. The energy demand has determinants that include previous load data,weather, time of day, economic
factors etc. An ANN machine learning model is utilised to forecast the total hourly load for 3
towns in Panama. The particular energy determinants for this dataset included hourly
temperature, humidity, precipitation, wind speed, nature of holiday and the electrical load
demand of the three towns from 2015 to 2020. The ANN model was developed, trained,
tested and validated using hourly data randomly attained from the open source dataset. The
model’s performance was evaluated at each level using methods such as MAPE, MSLE and
correlation of predicted data with the actual data. The built ANN model was deployed via a
web application in which the input is a CSV file containing the hourly data for the specified
expected energy demand determinants for a whole day. This input file includes weather
parameters such as humidity, liquid precipitation,temperature and wind as well as the day of
the week and whether it is a public holiday or not. The web application the provides an
interface that can be utilized by the utility or transmission company to determine the hourly
load forecast data for the next 24 hours of the day as well as the peak demand for that day.