Enhanced Domestic Short-Term Load Forecasting Using a Hybrid Deep Learning Technique

There are numerous challenges associated with operating today's electrical grids such as: shift in consumer preferences, the increased utilization of the renewable power sources and the need for load regulation. It remains important that way because without good electrical load management short-term load forecasting (STLF) helps in improving the effectiveness of the grid system. This paper seeks to make a comparison of two complex deep learning algorithms which are Bidirectional Long Short Term Memory (Bi-LSTM) and Convolutional Neural Network Bidirectional LSTM (CNN-BiLSTM) in the context of predicting consumption of energy in residential houses. The Bi-LSTM stands for the bidirectional LSTM; such an arrangement is more effective than typical LSTM displays as it makes temporal dependencies in both layers like as the backward and forward respectively. On the other hand, the CNN-BiLSTM model comprises of CNN layers used in extracting features of data and Bi-LSTM used in handling spatial and temporal features. For the purpose of an evaluation, the results of presented models have been compared with other conventional approaches by using the measures like Root Mean Squared Error (RMSE), Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE). The results found out that the CNN-BiLSTM model indicates improved precision and is better at predicting the energy usage patterns as compared to the Bi-LSTM and traditional predictive models. The paper also present the direction for the enhanced load forecast using the complex structure of neural networks and the future directions to work upon the improvement of load forecast in the energy management system.