A Cluster-Based Deep Learning Model for Energy Consumption Forecasting in Ethiopia

Abstract

The non-linearity and high variability of residential energy consumption data makes household energy prediction more challenging yet vitally important for efficient grid operation and power distribution scheduling. Neither traditional regression techniques nor conventional machine learning models are able to produce accurate forecasts for residential sector. Even though, deep learning methods are demonstrating remarkable performance in many complex problems including time series load forecasting. However, many issues need further investigation by integrating deep learning with clustering algorithms or hybridizing many deep learning models to obtain models with better accuracy. Hence, the aim of the proposed study is to investigate effectiveness of clustering methods to improve deep learning model for energy consumption prediction. In this paper, the K-means algorithm successfully identified interesting cluster groups and energy consumption patterns. Based on well-known error metrics, Long-Short Term Memory networks achieved greater accuracy on clustered data and outperformed the Gated Recurrent Units and Deep Feedforward Neural Network approaches.