The Effect of Input Length on Prediction Accuracy in Short-Term Multi-Step Electricity Load Forecasting: A CNN-LSTM Approach

This project investigates the effect of input sequence length on the accuracy of short-term multi-step electricity load forecasting using deep learning techniques. Using a real-world dataset from Panama's electricity consumption records, we implemented and compared the performance of four models: GRU, CNN-LSTM, Bidirectional LSTM (BiLSTM) with Dense layers, and a CNN-RNN hybrid. The models were evaluated for both hourly and daily forecasting to determine the optimal input configuration and architecture for high-precision predictions. Among these, the CNN-LSTM model demonstrated superior performance and was selected for deployment. To enable practical usage, we developed a Flask-based web application with an intuitive front end using HTML, CSS, and JavaScript. After user login, the application allows users to input specific dates and receive accurate load forecasts along with interactive graphs. This system provides a reliable and user-friendly platform for electricity demand forecasting, supporting better planning and decision-making in energy management.

Keywords:
Short-term load forecasting, CNN-LSTM, GRU, BiLSTM, CNN+RNN, electricity demand, input length, deep learning, Flask web app, time series prediction, energy forecasting.