Developing Electricity Energy Consumption Prediction Model Using Machine Learning: The Case of Bale Robe Town

Abstract   The electric industry is the backbone of the global energy sector and one of the most significant service providers. The only national company in our nation supplying electricity is Ethiopia Electric Utility. The dynamic and highly competitive character of the electric power sector means that companies must constantly keep up with the demands and desires of their customers, both individually and as organizations. The study was done on Ethiopian electric utility consumers in the Bale Robe district of the Oromia region, with data collected from January 2018 to January 2022. Based on data collected from Ethiopian electric utility Bale Robe Twon customers' electricity usage. I can create a prediction model on a daily, weekly, monthly, and yearly. In previous studies, a significant gap was observed between model data types and available data in Ethiopian power utilities. Most studies used limited datasets and outdated data mining tools, prompting the need for research using machine learning, time series, and hybrid algorithms. This research aims to develop a machine learning model to predict electricity consumption across different time frames. Data preprocessing techniques were applied to optimize artificial neural networks (ANNs) and genetic algorithms (GA), and all data were standardized to a common coordinate system for model alignment, addressing tasks like handling missing data and normalizing datasets   This research offers a hybrid of an ANN-GA model utilizing machine learning to analyze the volatility and variability of customers' electricity consumption.to construct the predictive model, the study used a variety of classification techniques and methodologies, including a hybrid of Artificial Neural Network and Genetic Algorithm, K-means Algorithm, logistic regressions (LRs), and SVM The results of the experiments showed that 97.4%,97%,97%,97%,97% respectively. The prediction model was built using a hybrid of artificial neural networks and GA, the goodness of fit data for the ANN-GA-based models ranged from January 2018 to 2022, with data from all Bale Robe utility customers included. The machine learning data set contains 18 attributes and 38,853 instances.   Therefore, based on the study's findings, the researcher would like to suggest that the electric industry carry out additional research and develop a system that allows for the best possible management of power usage and identify their demand.