A Short-Term Load Forecasting by Using Hybrid Model

Forecasting a region's energy needs, or “load,” involves calculating how much power will be required in the future. It often implies past data and considers various parameters such as weather, time of day, season, etc., that may affect energy consumption. Short-Term Load Forecasting (STFL) involves looking forward anywhere from twenty-four hours to a few weeks to estimate the power demand. It is mainly use for load balancing, unit commitment and economic dispatch operations. Therefore, in this paper, a novel approach consisting of Quantile Long Short-Term Memory (Q-LSTM) is presented for a short-term forecasting of the load data collected from the database of Electricity System Operator (ESO) for Great Britain. The validity of the suggested approach is confirmed by comparing the forecasted outcomes with those of well-established methods: LSTM and Convolutional Neural Network (CNN) with LSTM (CNN-LSTM) using Root Mean Square Error (RMSE) and Mean Absolute Error (MAPE). The findings express the superiority of the proposed system over LSTM and CNN-LSTM networks. Moreover, the RMSE and MAPE of Q-LSTM at quantile of 0.1 are 9.3% and 23.84% better than LSTM network, respectively. Similarly the recorded values of RMSE and MAPE of Q-LSTM at quantile of 0.1 are improved by 8.9% and 9.3%, respectively, as compared to CNN-LSTM network.