Integrated Prediction of Household Energy Consumption and Solar PV Generation Using Linear Regression and Hybrid CNN–LSTM

HRMARS - This study proposes an integrated framework for predicting household electricity consumption and solar photovoltaic (PV) generation by combining user behavior and weather data. Meteorological data from BMKG Banjarbaru and household survey data were utilized, incorporating behavioral variables such as appliance usage frequency, watt meter capacity, and household characteristics. A comparative analysis was conducted using multiple models, including Linear Regression, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree Regression (DTR), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM). In addition, a hybrid CNN–LSTM model was developed to enhance prediction performance for solar PV generation. Model evaluation was performed using MAE, RMSE, and MAPE under different data split scenarios (70:30, 80:20, and 90:10). The results show that the hybrid CNN–LSTM model achieves consistent and accurate performance in predicting solar PV generation, while Linear Regression provides stable and interpretable results for household energy consumption. From a behavioral perspective, electricity usage is primarily influenced by usage related factors rather than demographic characteristics. This study contributes by integrating demand side and supply side prediction into a unified framework for energy management. The findings reveal that energy supply is more predictable due to environmental factors, while energy demand is strongly influenced by user behavior. This integrated perspective provides practical insights for improving decision-making and supporting sustainable energy management.