Prediction of Electricity Consumption in Residential Areas using Temporal Fusion Transformer and Convolutional Neural Network

Abstract

The consumption of energy in the residential area causes adverse impacts on the environment. The mitigation or maintenance of power consumption can be the main step to preserve electricity for the future and proper supply. In context with this, the work focuses on predicting the consumption of energy with the novel hybrid tactic. The hybrid tactic is the integration of Temporal Fusion Transformer (TFT) and Convolutional Neural Network (CNN) (HTFT-CNN). This work is developed to predict the usage of energy across varying time frames with the grip of a multivariate time series of the power consumption of individual residential areas. The proposed HTFT-CNN is implemented to combine both the feature and temporal-based data and can be utilized to observe intricate consumption patterns. The Attention mechanism (AM) is implemented for the fusion of features that are obtained using the proposed HTFT-CNN tactic. The multi-step (k=24) for input sequences and k=24 is the length of input sequence at 24 hours. Simulations are conducted to analyze the robustness and forecasting accuracy of the designed model with the parameters such as Root Mean Square error (RMSE), and Mean Absolute Percentage Error (MAPE). The analyzed performance depicts that the proposed design can be used for planning and energy management in the residential area with minimized RMSE and MAPE values.