SISTEM PREDIKSI KONSUMSI ENERGI LISTRIK SUBSIDI DAN NON-SUBSIDI BERBASIS WEB MENGGUNAKAN METODE RECURRENT NEURAL NETWORK (RNN): STUDI KASUS KOTA SUKABUMI

Sukabumi City is experiencing a consistent increase in electricity demand each year across various sectors, including residential, industrial, and public facilities. Electricity consumption is categorized into two main types: subsidized and non- subsidized. However, the current distribution planning process is still performed manually and tends to be reactive, which limits effective decision-making in managing electrical loads. This study aims to develop a web-based electricity consumption prediction system using the Recurrent Neural Network (RNN) method, capable of generating detailed forecasts for each consumption category. The system was developed using the Laravel framework to ensure flexible access across devices and to support real-time monitoring and decision-making. The development process included collecting historical electricity consumption data in Sukabumi City, conducting a literature review, and applying Unified Modeling Language (UML) for system analysis and design. The application displays actual consumption data, predictive outputs, and interactive visualizations to help users better understand usage trends. Evaluation results show that the RNN model achieved a prediction accuracy of 95.98%, with a Mean Absolute Error (MAE) of 11.14 kWh, a Mean Squared Error (MSE) of 186.47 kWh², and a Root Mean Squared Error (RMSE) of 13.65 kWh. These results demonstrate the model’s high reliability in forecasting electricity usage based on historical patterns. Furthermore, black-box testing confirmed that all core features—including prediction, visualization, and user management—function correctly and meet the intended system requirements.
Keywords: Energy, Electricity, Prediction, Web, RNN.