A CNN and BiLSTM Fusion Approach Toward Precise Appliance Energy Forecasts

Abstract

Optimizing the energy grid management will enhance the effective and efficient use of generated energy. Accurate energy estimations allow power-generating firms to use dynamic energy management strategies to maintain stability across the smart grid. This paper presents a hybrid predictive modelling approach for forecasting the energy consumption of household appliances by combining 1D Convolutional Neural Networks (1D-CNN) with Bi-Directional Long Short-Term Memory (BiLSTM). The hybrid architecture combines CNNs' spatial feature extraction capabilities with BiLSTMs' sequential memory retention to comprehensively understand appliance use patterns. The proposed model is trained and validated on a dataset of household energy consumption, demonstrating superior performance compared to individual CNN or BiLSTM models. By incorporating both spatial and temporal data, energy consumption forecasts become more accurate and adaptable, making them highly appropriate for real-time applications and demand-side management. Utilizing the Analysis of Variance (ANOVA) F-measure feature selection technique and BiLSTM with a 1D-CNN hybrid deep learning model, the Root Mean Square Error (RMSE) of 1.745 and Coefficient of Determination (R 2 score) of 0.997.