Smart Technology and Artificial Intelligence
Optimizing Input Sequence for CNN-LSTM Multi-Step Load Prediction
Authors: AV. Sriharsha, Senji Lokanadha Reddy Punith, Kunchapu Nandusree, Maddila Ajay and Nukala Hanumath Rakesh
Publishing Date: 03-06-2026
ISBN: 978-81-975670-2-5
Abstract
Short-term electricity load forecasting helps sustain grid operations while increasing power generation output and supporting efficient energy distribution. The study examines how different input sequence lengths affect deep learning model performance during multi-step load forecasting. The researchers conducted a study to assess four models which included Gated Recurrent Unit (GRU) and CNN-LSTM and Bidirectional LSTM (BiLSTM) and CNN-RNN through testing on a real-world electricity consumption dataset from Panama which included both hourly and daily forecasting tasks. Tested multiple historical window sizes to study input sequence optimization effects while they used Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) to assess model performance. The results show that CNN-LSTM model delivers better accuracy for daily forecasts while BiLSTM model produces the most precise hourly forecasts. The research shows that using the optimal input sequence length leads to better forecasting accuracy because it helps predict time-based patterns while reducing unnecessary data. The research investigates historical load data together with time-related features but it recognizes that external elements like weather and holidays significantly affect electricity demand which should be examined in upcoming research. The team created a Flask-based web application which allows users to predict and display real-time load data through the most successful model.
Keywords
Short-term load forecasting, CNN-LSTM, GRU, BiLSTM, CNN-RNN, electricity demand, input sequence length, deep learning, web application.
Cite as
AV. Sriharsha, Senji Lokanadha Reddy Punith, Kunchapu Nandusree, Maddila Ajay and Nukala Hanumath Rakesh, "Optimizing Input Sequence for CNN-LSTM Multi-Step Load Prediction", In: Mukesh Saraswat, Sandeep Kumar, Manjunatha Sughaturu Krishnappa and Rakesh Keshava (eds), Smart Technology and Artificial Intelligence, SCRS, India, 2026, pp. 1-11. https://doi.org/10.56155/978-81-975670-2-5-1