Analysis of CNN-LSTM based hybrid Model for Improving Individual Household Load Forecasting

In recent years, time series forecasting has become one of the most important areas of study in electrical demand forecasting, particularly for the optimal and predictable operation of power grids. In many countries, non-renewable energy sources such as thermal and hydro power plants still play a major role in electricity generation. For the smooth operation of these plants, accurate load prediction is crucial. To generate electrical power efficiently, it is essential to determine its demand in advance. Real-time analysis of electricity demand has become possible thanks to smart metering technology.Short-Term Load Forecasting (STLF) using advanced forecasting techniques provides near real-time predictions of electricity demand. This paper primarily focuses on comparing different artificial neural network (ANN)-based forecasting models for hourly electricity demand forecasting at the individual household level. Among these models, Long Short-Term Memory (LSTM) networks, especially when used in hybrid approaches, show significant improvements in STLF performance. The models are evaluated based on forecasting errors, specifically RMSE.