Short-Term Electrical Load Prediction for Future Generation Using Hybrid Deep Learning Model

Power generation is increasing worldwide every year to cope with ever-increasing energy demand. Therefore, a significant necessity exists for forecasting the load demand to manage and increase electricity production capacity. Short-term load forecasting (STLF) using artificial neural network has become one of the most efficient and widely popular methods. This paper proposes a hybrid network of Long Short-Term Memory (LSTM) network and Convolutional Neural Network (CNN) to predict demand for seven days into the future. The proposed CNN-LSTM method is compared with various deep learning techniques such as vanilla neural network and gated recurrent unit (GRU). Power Grid Company of Bangladesh (PGCB) has the responsibility of reliable power transmission all over the country. Each model is trained and tested on multivariate historical data collected from the daily report section of PGCB website for the Mymensingh Division in Bangladesh. Various input features such as temperature, peak generation at evening, maximum generation, month and the season of the year are used to aid the prediction. It is found that the proposed CNN-LSTM method outperforms the other models with a MAPE error rate of 2.8992%, which is less than the MAPE error of 5.5554% for demand estimation of seven days used by PGCB.