Indian Journal of Science and Technology
DOI: 10.17485/IJST/v17i11.3246
Year: 2024, Volume: 17, Issue: 11, Pages: 1016-1027
Original Article
Nilakanta Kshetrimayum1*, Khumukcham Robindro Singh1, Nazrul Hoque1
1Department of Computer Science, Manipur University, Indo-Brahma Road, Imphal, 795003, Manipur, India
*Corresponding Author
Email: [email protected]
Received Date:27 December 2023, Accepted Date:08 February 2024, Published Date:29 February 2024
Background: Short-term Load Forecasting (STLF) is vital for grid stability, ensuring a steady power supply and resource efficiency. However, the literature review underscores imperfections in current methods, emphasizing the necessity for additional research in this domain. Objectives: This study introduces an effective framework for multi-step STLF, enhancing predictive accuracy by integrating state-of-the-art DNN models like Long Short Term Memory (LSTM) and Convolutional Neural Network (CNN) in a hybrid architecture, leveraging their complementary strengths. Method : A parallel hybrid network consisting of LSTM and a Two-Dimensional Convolutional Neural Network (2D-CNN), which operates load sequence input and 2D converted load input, is employed. Their combined context vector is finally used for predicting the preceding 24 load sequences. Extensive comparative testing involves various existing STLF methods (Artificial Neural Network (ANN), LSTM Sequence-to-Sequence (LSTM-S2S), Gated Recurrent Unit with Genetic Algorithm (GRU-GA), CNN-LSTM, GRU-CNN, and Parallel LSTM and CNN Network (PLCNet)) experimented across nine load datasets, using Mean Absolute Percentage Error (MAPE) as the evaluation metric. Findings: The proposed model demonstrates enhanced MAPE, with values of 3.34, 5.67, 4.92, 4.84, 5.24, 4.73, 5.15, 5.49, and 3.96 across nine datasets. The critical distance diagram further validates these results. The findings from this comparative analysis underscore the efficacy of the proposed multi-step STLF method, showcasing significant improvements in forecasting accuracy. Novelty: The presented network architecture is novel in its fusion of LSTM and 2D-CNN networks through the Gramian Angular Field (GAF) technique. Additionally, the parallel operation of LSTM and 2D-CNN generates distinct context vectors combined to form the final output.
Keywords: Parallel CNNLSTM, Gramian Angular Field, Electricity Demand Prediction, Short-term Load Forecasting, Multi-step Forecasting
© 2024 Kshetrimayum et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Published By Indian Society for Education and Environment (iSee)
Subscribe now for latest articles and news.