Electric demand Prediction using LSTM for Medium Term Planning

In this research, the future projection of electric demand has been carried out using a deep learning approach. Specifically, a deep learning network was adopted to enhance the precision of the forecasts. Historical electric demand data were utilized as a foundation for the analysis, ensuring that the model leverages long-term trends and patterns. The data spans from 1990 to 2024, with records from 1990 to 2020 used for training the model and the years 2021 to 2024 designated for testing and validating the predictions against actual observations. Root Mean Square Error (RMSE) was used in this study as the fitness function to assess how accurate the predictions were. This metric calculates the error between the actual historical data and the projected estimates, ensuring a reliable evaluation of model performance. Additionally, the Long Short-Term Memory (LSTM) algorithm was adopted in this study to estimate electric demand. LSTM, a powerful recurrent neural network (RNN) architecture, is particularly ideal for prediction of time-series and data sequences. By capturing dependencies over long time horizons, the algorithm enhances the accuracy of the predictions. The research extends its electric demand projections up to the year 2030. These projections have been validated by comparing them with the estimates provided in the National Electricity Plan of India, ensuring that the results align with national planning and energy policies. This study not only demonstrates the effectiveness of deep learning models like LSTM in electric demand forecasting but also provides a valuable tool for future energy management and decision-making.